Patent Grant
6909724
The present invention relates generally to providing network services such as load balancing, packet filtering, or Network Address Translation (NAT). Network services are provided using service managers that synchronize instructions that are carried out using a distributed network of forwarding agents.
As the IP protocol has continued to be in widespread use, a plethora of network service appliances have evolved for the purpose of providing certain network services not included in the protocol and therefore not provided by standard IP routers. Such services include NAT, statistics gathering, load balancing, proxying, intrusion detection, and numerous other security services. In general, such service appliances must be inserted in a network at a physical location where the appliance will intercept all flows of interest for the purpose of making its service available.
Network service appliance 130 provides a network service such as load balancing, caching, or security services. In providing security services, network service appliance 130 may function as a proxy, a firewall, or an intrusion detection device. For purposes of this specification, a network service appliance that acts as a load balancer will be described in detail. It should be noted that the architecture and methods described are equally applicable to a network service appliance that is functioning as one of the other above described devices.
Network service appliance 130 is physically located between the group of servers and the clients that they serve. There are several disadvantages to this arrangement. First, it is difficult to add additional network service appliances when the first network service appliance becomes overloaded because the physical connections of the network must be rerouted. Likewise, it is difficult to replace the network service appliance with a back up network service appliance when it fails. Since all packets pass through the network service appliance on the way to the servers, the failure of the network service appliance may prevent any packets from reaching the servers and any packets from being sent by the servers. Such a single point of failure is undesirable. Furthermore, as networks and internetworks have become increasingly complex, multiple services may be required for a single network and inserting a large number of network service appliances into a network in places where they can intercept all relevant packet flows may be impractical.
The servers may also be referred to as hosts and the group of servers may also be referred to as a cluster of hosts. If the group of servers has a common IP address, that IP address may be referred to as a virtual IP address (VIPA) or a cluster address. Also, it should be noted that the terms client and server are used herein in a general sense to refer to devices that generally request information or services (clients) and devices that generally provide services or information (servers). In each example given it should be noted that the roles of client and server may be reversed if desired for a particular application.
A system that addresses the scalability issues that are faced by network service appliances (load balancers, firewalls, etc.) is needed. It would be useful to distribute functions that are traditionally performed by a single network element and so that as much function as possible can be performed by multiple network elements. A method of coordinating work between the distributed functions with a minimum of overhead is needed.
Although network service appliances have facilitated the development of scalable server architectures, the problem of scaling network service appliances themselves and distributing their functionality across multiple platforms has been largely ignored. Network service appliances traditionally have been implemented on a single platform that must be physically located at a specific point in the network for its service to be provided.
For example, clustering of servers has been practiced in this manner. Clustering has achieved scalability for servers. Traditional multiprocessor systems have relatively low scalability limits due to contention for shared memory and I/O. Clustered machines, on the other hand, can scale farther in that the workload for any particular user is bound to a particular machine and far less sharing is needed. Clustering has also facilitated non-disruptive growth. When workloads grow beyond the capacity of a single machine, the traditional approach is to replace it with a larger machine or, if possible, add additional processors within the machine. In either case, this requires downtime for the entire machine. With clustering, machines can be added to the cluster without disrupting work that is executing on the other machines. When the new machine comes online, new work can start to migrate to that machine, thus reducing the load on the pre-existing machines.
Clustering has also provided load balancing among servers. Spreading users across multiple independent systems can result in wasted capacity on some systems while others are overloaded. By employing load balancing within a cluster of systems the users are spread to available systems based on the load on each system. Clustering also has been used to enable systems to be continuously available. Individual application instances or machines can fail (or be taken down for maintenance) without shutting down service to end-users. Users on the failed system reconnect and should not be aware that they are using an alternate image. Users on the other systems are completely unaffected except for the additional load caused by services provided to some portion of the users that were formerly on the failed system.
In order to take full advantage of these features, the network access must likewise be scalable and highly available. Network service appliances (load-balancing appliances being one such example) must be able to function without introducing their own scaling limitations that would restrict the throughput of the cluster. A new method of providing network services using a distributed architecture is needed to achieve this.
When network services are provided using a distributed architecture, the problem of synchronizing actions among the forwarding agents arises. Traditional two-phase commit protocols can be used, but such protocols are time consuming and also use considerable resources for the purpose of generating confirmations. This is especially problematic for a network that includes a number of routers or switches having forwarding agents and where each router or switch must synchronize service instructions with other devices. (It should be noted that for the purpose of brevity, routers will be referred to in examples hereinafter; it should be understood that a switch may also be used where appropriate.) Traditional two-phase commit schemes are too cumbersome and slow for the routing fabric. Since the routers are separate devices physically located remotely from each other, a scheme relying on a shared media used by the routers would be impractical. Each service manager would use resources to confirm every instruction to every router.
What is needed is a scheme for synchronizing instructions across multiple routers in the absence of shared media or two-phase commit protocols.
A scheme wherein a service manager is used as a point of synchronization and distribution of filters among routers is disclosed. Filters are used to determine how routers are to handle packets. Forwarding agents on the routers cache filters received from the service manager. Instructions are sent to the forwarding agents without reliance on a two phase commit scheme to confirm receipt of instructions. Instructions may be periodically resent on an as needed basis. In this robust scheme, forwarding agents consult the service manager whenever instructions are not found by the forwarding agent for handling packets controlled by the service manager. Forwarding agents only consult the service manager when necessary and the scheme does not require verified communication between the service manager and the forwarding agent.
It should be appreciated that the present invention can be implemented in numerous ways, including as a process, an apparatus, a system, a device, a method, or a computer readable medium such as a computer readable storage medium or a computer network wherein program instructions are sent over optical or electronic communication lines. Several inventive embodiments of the present invention are described below.
In one embodiment, a method of obtaining instructions for routing a packet includes receiving a packet having a packet flow identifier that includes a packet source IP address, a packet destination IP address, a packet source port, and a packet destination port and checking whether the packet flow identifier matches a stored instruction. If the packet flow identifier does not match a stored instruction, whether the packet flow identifier matches a stored criteria is checked and if the packet matches a stored criteria, the packet is forwarded to a service manager.
In another embodiment, a forwarding agent includes a packet receiving interface configured to receive a packet having a packet flow identifier that includes a packet source IP address, a packet destination IP address, a packet source port, and a packet destination port and a processor configured to check whether the packet flow identifier matches a stored instruction and if the packet flow identifier does not match a stored instruction, checking whether the packet flow identifier matches a stored criteria. A packet forwarding interface forwards the packet to a service manager if the packet matches a stored criteria.
In another embodiment, a network service manager includes a set up interface for specifying a set of packets for which the service manager will provide a network service and a forwarding agent communication interface on which packet interest instructions can be multicast to a plurality of forwarding agents for the purpose of instructing the plurality of forwarding agents to send certain packets in the set of packets to the service manager. On a packet receiving interface, the service manager can receive packets sent from a reporting forwarding agent in response to the multicast packet interest instructions. A processor processes the packet received on the packet receiving interface and determines an action for the reporting forwarding agent to take regarding the packet.
In another embodiment, a method of processing a packet at a router includes receiving a packet having packet flow identifier that includes a packet source IP address, a packet destination IP address, a packet source port, and a packet destination port and checking whether the packet flow identifier matches a general filter. Whether the packet flow identifier matches a specific filter is checked and if the packet flow identifier does not match the specific filter but does match the general filter the packet is sent to a service manager indicated by the general filter.
In another embodiment, a method of synchronizing packet handling instructions from a service manager to a router includes sending an interest filter to the router, the interest filter specifying a plurality of flows for which packets are to be sent to the service manager and receiving a requested packet corresponding to a specific one of the flows. Actions to be applied to the specific one of the flows are determined and an action filter is sent to the router that specifies an action to be performed for each packet in the specific one of the flows.
These and other features and advantages of the present invention will be presented in more detail in the following specification of the invention and the accompanying figures which illustrate by way of example the principles of the invention.
The present invention will be readily understood by the following detailed description in conjunction with the accompanying drawings, wherein like reference numerals designate like structural elements, and in which:
A detailed description of a preferred embodiment of the invention is provided below. While the invention is described in conjunction with that preferred embodiment, it should be understood that the invention is not limited to any one embodiment. On the contrary, the scope of the invention is limited only by the appended claims and the invention encompasses numerous alternatives, modifications and equivalents. For the purpose of example, numerous specific details are set forth in the following description in order to provide a thorough understanding of the present invention. The present invention may be practiced according to the claims without some or all of these specific details. For the purpose of clarity, technical material that is known in the technical fields related to the invention has not been described in detail in order not to unnecessarily obscure the present invention.
In the example shown, forwarding agent 231 is connected to server 221 and server 222 and forwarding agent 232 is connected to server 222 and server 223. Thus, server 222 may communicate with network 210 through either of the forwarding agents, server 221 communicates with network 210 exclusively through forwarding agent 231, and server 223 communicates with network 210 exclusively through forwarding agent 232. This arrangement may be generalized to include an arbitrary number of servers connected to an arbitrary number of forwarding agents with individual servers connected to arbitrary subsets of the forwarding agents.
A service manager 241 and a second service manager 242 also communicate with the forwarding agents. The service managers provide the decision making capability that is required to provide a network service such as load balancing. The service managers send specific instructions to each of the forwarding agents detailing how certain flows of packets are to be processed. Such packet processing may include simply routing the packet, gathering statistics about the packet, sending the packet to a service manager, sending a notification that the packet has been seen to a service manager, modifying the packet, or using a special method such as tunneling or tag switching to send the packet to a destination other than the destination specified by the destination IP address included in the packet header. It should also be noted that forwarding agents in other embodiments also modify other aspects of packets, including packet source and destination addresses and port numbers and, in some instances, packet data.
The service managers communicate with the forwarding agents to give the agents instructions relating to how to handle packets for various flows that are routed through the forwarding agents. It is useful at this point to review certain terminology used herein relating to connections and flows.
As used in this specification, a connection consists of a set of flows. A flow is a set of related packets sent between two end stations. A flow may be identified with layer 3 and layer 4 parameters, depending on the protocol being used. For example, for TCP and UDP, a flow is identified by five parameters: the source and destination IP addresses and port numbers and the protocol. For ICMP, flows are defined by three parameters: the source and destination IP addresses and the protocol.
TCP connections will be described in detail in this specification. It should be appreciated that the techniques disclosed apply to other types of connections as well. TCP connections are defined by a 5-tuple that includes the source and destination IP addresses, the source and destination port numbers, and an identification of the protocol that applies to the packet. The source and destination IP addresses and ports for packets going in one direction between the devices are reversed for packets going in the opposite direction. That is, when the direction that a packet is travelling is reversed, the source becomes the destination and the destination becomes the source. Packets flowing in one direction of a connection are in the same flow.
A connection transfers data between applications on two machines having IP addresses and the applications, correspond to port numbers. If the protocol is set by convention to be a certain protocol such as TCP, then a protocol identifier may not be required. The 4 remaining numbers, the source and destination IP addresses, and the source and destination port numbers, are sometimes referred to as a quad. In this specification, the 5-tuple that includes the source and destination IP addresses, the source and destination port numbers and a protocol identification will be referred to as an affinity key. Each unique affinity key thus defines a flow in one direction of a connection. If the source and destination IP addresses and port numbers are reversed for a single affinity key, then it becomes an affinity key that corresponds to a flow in the opposite direction for the same connection. In general, a flow may be identified by a source IP address and destination IP address, by a source IP address, destination IP address and protocol, by a quad, by an affinity key 5-tuple, by only a source and destination IP address or by other information available in a packet header. The term, “flow identifier” is intended to refer to any such method of identifying a flow.
Affinity keys are used by the service managers to identify flows passing through forwarding agents which are to be handled by the forwarding agents in a certain manner. Forwarding agents can accomplish their required tasks with only limited processing capability. Forwarding agents need not determine how to handle certain flows or make decisions such as load balancing or security decisions relating to the flows. The service manager performs those functions and forwards specific instructions to forwarding agents detailing exactly what actions are to be taken for each flow. Instructions for how to handle packets are specified for each flow by the service managers using an affinity key. A specific affinity key that is sent to a forwarding agent together with instructions detailing how packets for flows specified by the affinity key are to be handled is referred to as a fixed affinity.
In addition to specifying instructions for each flow, service managers must also obtain information about each new flow from the forwarding agents. For example, when a service manager provides load balancing through a set of forwarding agents, the service manager uses fixed affinities to provide specific instructions to the forwarding agents detailing where packets for each load balanced flow are to be forwarded. In addition to providing those specific instructions, the service manager also provides general instructions to each forwarding agent that specify which new flows the service manager is interested in seeing. These general instructions are provided using wildcard affinities. Wildcard affinities, which are described in detail below, specify sets of flows that are of interest to a service manager. In one embodiment, this is done by specifying subnet masks that determine sets of source and destination IP addresses that will be forwarded to a service manager. In addition, ports or sets of ports and protocol may be specified in wildcard affinity as well. As is described further below, the use of wildcard affinities enables separate service managers to be configured to provide services for different sets of flows. Each service manager specifies the flows of interest to it and other service managers handle other flows. In this manner, service managers can be configured in parallel to share load.
Thus, service managers use wildcard affinities to specify flows for which they may be providing service and forwarding agents transfer packets for new flows to the appropriate service manager. Once a service manager determines how a certain flow is to be handled, the service manager sends a fixed affinity to each forwarding agent. The fixed affinity overrides the wildcard affinity stored in the forwarding agent that instructs the forwarding agent to forward packets to the service manager with specific instructions for the specific flow specified by an affinity key in the fixed affinity.
In the case of load balancing, service managers send wildcard affinities to forwarding agents. The wildcard affinities specify destination IP addresses that correspond to virtual IP addresses of server clusters that are to be load balanced by the service manager. The forwarding agents then forward new packets sent to those virtual IP addresses to the appropriate service manager. The service manager selects a server from the server cluster and then the service manager sends a fixed affinity to each forwarding agent that instructs the forwarding agent to forward packets for that specific flow to the selected server in the cluster. Forwarding agents may also forward packets for purposes other than load balancing. Packets may be forwarded to real IP addressees as well as virtual IP addresses.
In one embodiment, each forwarding agent is implemented on a router. In other embodiments, forwarding agents may be implemented on switches or other network devices and may be implemented on a coprocessor in a device that also performs another network function. When implemented on a router, the power of this architecture becomes clear. By infusing each router with a limited functionality provided by the forwarding agent, the service managers are able to provide network services without physically being inserted at the various points in the network where those services must be provided. The physical presence of each of the routers at those points is sufficient to enable network services to be provided. This contradicts the conventional wisdom regarding the restriction that all traffic inbound for a server cluster must pass through a single load-balancing engine. The combination of fast forwarding agents (be they ‘routers’ or IP-aware ‘switches’) and service managers (to provide synchronization and control) eliminates the scalability limitations of the past.
This specification will refer in detail to forwarding agents implemented on routers for the purpose of example. It should be remembered that forwarding agents may also be implemented on other devices and that the same or similar advantages may be realized.
The service managers send wildcard affinities to each of the forwarding agents that direct the forwarding agents to process packets that match the wildcard affinities in a certain manner. For example, a service manager may request to be notified when certain packets are received by the routers that include the forwarding agents. When a packet that matches such an instruction is received, the forwarding agent notifies the service manager and the service manager determines what to do with that packet and future packets for the flow based on the network service being provided. Instructions are then sent from the service manager to the forwarding agent at the router that allow the router to process the packets in accordance with the decisions made by the service manager.
In addition to specifying that a service manager is to be notified upon receipt of a certain type of packet, wildcard affinities may also specify other actions to be taken. For example, a wildcard may specify an IP address to which packets are to be forwarded without notification to the service manager. Packets may also be copied to a service manager or other device and packets may also be denied or dropped.
It should be noted that the service managers also may be connected to one or more of the servers and may in some cases forward packets received from forwarding agents or received from the network directly to certain servers. However, it is significant that the service managers need not be connected to servers for which they are managing packet traffic. The service manager may accomplish all packet routing through forwarding agents by sending instructions to forwarding agents. It should also be noted that the service managers may also be connected to each other for the purpose of coordinating their instructions or providing backup services.
Forwarding agent 250 also includes a service manager interface 258 that allows packets to be sent to and received from a service manager. In addition, the service manager interface allows service managers to send fixed and wildcard affinities to the forwarding agent. In one embodiment, a separate interface is used for the purpose of sending wildcard affinities to forwarding agents using multicast. In other embodiments, a single interface may be provided between the service manger and the forwarding agent. The forwarding agent also includes a network interface 260 that is used to send and receive packets to and from other devices on the network.
It should be noted that the network interface and the service manager interface may be the same interface in certain embodiments. In such embodiments, all communication between the forwarding agent and the service manager is carried on the same network as packets processed by the forwarding agent.
A forwarding agent may be implemented on various network devices. A forwarding agent may be implemented on a network device dedicated to acting as a forwarding agent but the true power of the system is realized when forwarding agents are implemented on network devices that already are included in a network for some other purpose. Forwarding agents may be implemented on routers that already exist at strategic points in a network for intercepting packets and providing a service using a forwarding agent.
Service manager 270 also optionally includes a network interface 278. Network interface 278 allows the service manager to directly forward packets into the network for which it is providing a service. If no network interface is provided, then the service manager can still forward packets by sending them to a forwarding agent.
A forwarding agent interface 280 is included on the service manager for the purpose of allowing the service manager to send packets and affinities to forwarding agents. Forwarding agent interface 280 may include more than one interface. For example, in one embodiment, a separate interface is used for multicasting wildcard affinities to all forwarding agents and a separate interface is used for the purpose of unicasting fixed affinities to individual forwarding agents and forwarding packets to individual forwarding agents.
Service manager 270 may also include a service manager interface 282 used to communicate with other service managers. The service manager may communicate with other service managers for the purpose of providing a fail over scheme of backup service managers. Operational status of service managers may be communicated on the service manager interface and a master service manager may send configuration information about flows being supported through backup service managers so that the backup service managers can function in place of the master service manager should it fail.
A service manager may be implemented on a standard microcomputer or minicomputer. In one embodiment a service manager is implemented on a UNIX workstation. A Service manager may also be implemented on other platforms including Windows, an embedded system or as a system on a chip architecture. A service manager also may be implemented on a router.
One network service that can be readily provided using the architecture described in
As an example, client 304 may wish to establish a TCP connection with a virtual machine having a virtual IP address. It should be noted that other types of connections may also be established. To establish the TCP connection, client 304 sends a SYN packet with a destination address corresponding to the virtual IP address. The SYN packet is received by forwarding agent 302. Forwarding agent 302 determines that the destination address of the SYN packet matches the wildcard affinity broadcast by service manager 300. The action included in the broadcast wildcard affinity specifies that all packets matching the wildcard affinity are to be forwarded to the service manager. Therefore, forwarding agent 302 forwards the SYN packet to service manager 300.
Service manager 300 receives the SYN packet from the forwarding agent. It should be noted that, in one embodiment, forwarding agent 302 encapsulates the SYN packet in a special system packet when the SYN packet is sent to the service manager. Service manager 300 receives the SYN packet and processes the packet according to whatever service or services are being provided by the service manager. In the example shown, service manager 300 is providing load balancing between a first host 306 and a second host 308. Together, host 306 and host 308 comprise a virtual machine that services the virtual IP address that is the destination of the SYN packet sent by client 304. Service manager 300 determines the host that is to receive the SYN packet and that is to handle the connection initiated by the SYN packet. This information is included in a fixed affinity. The SYN packet is encapsulated with the fixed affinity and sent back to forwarding agent 302.
The fixed affinity sent to the forwarding agent 302 may include an action that directs the forwarding agent to dispatch the SYN packet directly to host 306. The action included in the fixed affinity may also direct the forwarding agent to translate the destination address of the packet to the IP address of host 306 and the packet may be routed to host 306 via one or more hops. In addition, as described below, tag switching may also be used to send the packet to the host that is selected by the service manager using its load balancing algorithm.
Thus, the SYN packet is directed to the host selected by service manager 300 without service manager 300 being inserted into the path of the packet between the hosts which comprise virtual machine 310 and client 304. The service manager broadcasts a wildcard affinity to all forwarding agents potentially in that path and the forwarding agents forward SYN packets to the service manager whenever a client establishes a new connection. The service manager then returns the SYN packet with a fixed affinity that directs the forwarding agent how to forward that SYN packet as well as future packets sent in the flow from the client to the virtual machine. The forwarding agent then sends the SYN packet on to the selected host using network address translation (NAT), tag switching, or some other method.
Thus,
The SYN packet is forwarded to the service manager and the service manager returns the SYN packet to the forwarding agent along with a fixed affinity that includes an action which specifies how the forwarding agent is to handle the SYN packet. When a SYN ACK is returned by the host, the forwarding agent again finds a wildcard affinity match and forwards the SYN ACK packet to the service manager. The service manager returns the SYN ACK packet to the forwarding agent along with a second fixed affinity that instructs the forwarding agent how to handle packets in the flow back from the host the client.
The first fixed affinity from the service manager includes an affinity key that corresponds to the flow from the client to the host and the second fixed affinity sent form the service manager to the forwarding agent contains an affinity key that corresponds to the flow from the host back to the client. Future packets in either flow sent from the client or the host match the affinity key in one of the fixed affinities and are handled by the forwarding agent according to the action contained in the fixed affinity. It is no longer necessary to forward such packets to the service manager. In some applications, the forwarding agent may continue to forward data about the packets to the service manager so that the service manager can monitor connections or maintain statistics about network traffic.
The examples shown in FIG. 3A through
In general, the service managers may communicate amongst themselves and with the forwarding agents in any manner appropriate for a particular system. The forwarding agents each are connected to a first server 430, a second server 432 and other servers up to an nth server 440. These servers may represent one or more virtual machines. Packets from the clients may be routed through either forwarding agent 410 or forwarding agent 412. In fact, packets corresponding to the same connection or flow may be routed at different times through different forwarding agents. To cope with this situation, the service managers multicast wildcard affinities to both forwarding agents. When either forwarding agent first receives a packet for a flow, that forwarding agent forwards the packet to the manager that has requested the packet using a wildcard affinity so that the service manager can provide the forwarding agent with the fixed affinity that defines how to handle the packet.
Host 1 receives the SYN packet from forwarding agent 1 and returns a SYN ACK packet back to client 500. However, for some reason, the SYN ACK packet from host 1 is routed not through forwarding agent 502, but instead through forwarding agent 512. Forwarding agent 512 receives the SYN ACK and notes that it matches a wildcard affinity corresponding to the flow of packets from host 506 to client 500. Forwarding agent 512 encapsulates the SYN ACK packet and sends it to service manager 504. Service manager 504 defines an action for the SYN ACK packet and includes that action in a second fixed affinity which it sends along with the encapsulated SYN ACK packet back to forwarding agent 512. Forwarding agent 512 then sends the SYN ACK packet on to client 500 where it is processed.
At this point, forwarding agent 502 has a fixed affinity for the flow from client 500 to the hosts and forwarding agent 512 has a fixed affinity for the flow from the hosts back to client 500. Each forwarding agent continues to handle flows without fixed affinities using the wildcard affinities. The service manager acts as a point of synchronization between the forwarding agents when the forwarding agents handle common flows.
Client 500 then sends a data packet which happens to be routed through forwarding agent 512 and not forwarding agent 502. Forwarding agent 502 has received the fixed affinity that provides instructions on how to deal with packets in the flow from client 500 to virtual machine 510. However, forwarding agent 512 has not yet received that fixed affinity. Forwarding agent 512 has received a wildcard affinity previously multicast by the service manager. Therefore, forwarding agent 512 detects a wildcard affinity match for the data packet and encapsulates the data packet and sends it to service manager 504.
Service manager 504 receives the data packet and notes that the data packet matches the previously defined first fixed affinity which was sent to forwarding agent 502. Service manager therefore does not run the load balancing algorithm again to determine where to route the data packet, but instead returns the first fixed affinity to forwarding agent 512 along with the data packet. Forwarding agent 512 receives the data packet and the fixed affinity and then has the same instructions as forwarding agent 502 for handling that data packet and other packets in the flow from client 500 to virtual machine 510. Forwarding agent 512 therefore translates the destination IP address of the data packet to the IP address of host 506 and forwards the packet on to host 506.
Thus, as long as wildcard affinities are received by each forwarding agent, the service manager is able to provide fixed affinities to each forward agent whenever a fixed affinity is required to provide instructions to handle packets for a given flow. Once a fixed affinity is defined for a flow, the same fixed affinity is provided to any forwarding agent that returns a packet to the service manager as a result of a wildcard match.
To provide a load balancing service for HTTP, a service manager sends a pair of wildcard affinities (one for each direction of flow to and from a virtual machine) to a multicast group that includes each available router in a network. The wildcard affinities specify a protocol and also indicate an exact match on the IP Address and HTTP port number for the virtual machine and an IP address and mask combination that identifies the client population that is serviced by the service manager. The client population serviced by the service manager is referred to as the client domain of the service manager. If multiple service managers are used, then each service manager may be configured to service a different client domain.
For example, if the majority of traffic is coming from a small number of firewalls, whereby the same foreign IP address is shared by many different clients, all those affinities can be assigned by one service manager. Thus, traffic from large sites can be isolated from other traffic and assigned to a different service manager.
Thus, the architecture is scalable and service managers may be added to handle client domains as needed. The set of clients serviced by each service manager can be changed by canceling the wildcards that each service manager has broadcast to forwarding agents and sending new wildcards specifying the new client domain.
When multiple service managers are included, it is important that the client domains specified by service managers performing the same service do not overlap. The task of assigning affinities for each client domain is centralized by the service manager serving that domain so all packets for a given flow are controlled by a single service manager. For example, if duplicate SYN packets are sent by a client, both should be directed to the same service manager and assigned the same fixed affinity. If the packets were directed to different service managers, then the service manager load balancing algorithms might assign different real machines to handle the connections as a result of the network being in a different state when the second SYN packet arrived. In addition, UDP unicasts from the same client must be assigned the same affinity and related connections (e.g., FTP control and data connections) must be assigned the same affinity.
Once the forwarding agents have received fixed affinities, packets intercepted that match a fixed affinity are processed as instructed in the set of actions specified in the fixed affinity. If a matching fixed affinity is not found, the packet is compared against the wildcard affinities to find manager(s) that are interested in this type of packet. If no appropriate Wildcard Affinity is found, normal IP routing occurs. Generally, a manager uses the wildcard affinity to be informed of flows it may be interested in. Once a manager has determined how a flow should be handled, it usually sends a fixed affinity so that the processing of subsequent packets for that flow can be offloaded to the forwarding agent. In some cases actions for certain flows can be predetermined by the service manager without seeing packets from the flow. In such cases, the actions may be specified in a wildcard affinity and no message need be sent to the service manager and no fixed affinity need be generated. The service manager may specify that it is still to receive certain packet types after a fixed affinity is sent by including an optional action interest criteria message segment with the fixed affinity.
In the load-balancing case, a fixed affinity is used to identify the server that is to receive this particular flow whereas a wildcard affinity is used to define the general class of packets for which load balancing is to be performed (all those matching the cluster address and port number for the clustered service) and to identify the manager that is to make the balancing decision for flows that match the wildcard affinity.
Fixed Affinities
IP Header:
It should be noted that if the protocol being used is not TCP or UDP, then the ports in the affinity key may be set to 0.
Fixed affinity 600 includes an affinity key 602. In addition, fixed affinity 600 contains information that dictates how a forwarding agent is to process packets that match the affinity key, and how the forwarding agent is to manage the affinity.
A dispatch flag 604 indicates whether the packet is to be dispatched to the forward IP address included in the fixed affinity. Setting the dispatch flag indicates that the packet is to be forwarded to a forward IP address 608 that is provided in the fixed affinity. The difference between dispatched and directed traffic is that dispatch traffic is forwarded directly from a forwarding agent to a specific server without translating the destination IP address of the packet. In other words, if a packet is dispatched, then the packet destination address is not used to forward the packet. Instead, a forwarding address contained in an affinity is used to forward the packet. If the connection is not dispatched but directed by the forwarding agent, then the packet IP destination must be translated using NAT if the packet is redirected to a specific server.
If forward IP address 608 is zero, then the packet is dropped after processing statistics as indicated by an information flag 606. Not setting the dispatch flag indicates that the packet is to be forwarded based on the address provided in the packet IP header.
Information flag 606 indicates whether or not statistics are to be gathered for packets forwarded using the fixed affinity. If the Information flag is set, statistics are updated for the forward IP address. In one embodiment, the statistics kept include:
Statistics for packets and bytes matching the affinity may be kept regardless of the setting of the Information flag.
Fixed affinity 600 also includes a time to live 610. Time to live 610 specifies the number of seconds before the fixed affinity should be timed-out from a fixed affinity cache maintained by a forwarding agent. If a time to live of 0 is specified, then that means that the fixed affinity is not to be cached by a forwarding agent and if a copy of the fixed affinity is already in the cache, it should be removed. Thus, service managers may remove fixed affinities that they have sent to forwarding agents by simply sending copies of those fixed affinities to the forwarding agents with time to live set to 0.
Each fixed affinity sent by a service manager is correlated to a wildcard affinity previously sent by the service manager. If a forwarding agent receives a fixed affinity for which no supporting wildcard affinity is found, the forwarding agent ignores the fixed affinity and discards it.
Wildcard Affinities
The netmasks and the source and destination IP addresses are used to specify ranges of addresses covered by the wildcard affinity. The source netmask is ANDed with the source IP address in the wildcard affinity. The source netmask is also ANDed with the source IP address from the packet. If the results of the two operations are equal, then the source IP address of the packet is considered to be in range of the wildcard affinity. Likewise, the destination netmask is ANDed with the destination IP address in the wildcard affinity. The destination netmask is also ANDed with the destination IP address from the packet. If the results of the two operations are equal, then the destination IP address of the packet is considered to be in range of the wildcard affinity. If both the source and the destination IP addresses of the packet are in the range of the wildcard affinity, and the ports and protocols also match, then the packet is said to match the wildcard affinity. It should also be noted that, in one embodiment, a zero specified for a port or a protocol matches all ports or protocols.
It should be noted that in other embodiments, other methods of specifying ranges for the wildcard affinity are used. For example, in one alternative arrangement, ranges of IP addresses are specified by specifying lower bound and upper bound IP addressees. All addresses between the two bounds fall within the range of the wildcard affinity. In some applications, multiple ranges may be specified. The method described above is particularly useful for specifying a single address, specifying all addresses in a subnet, or specifying every even or odd address, every fourth address, every eighth address, etc.
For example, to specify a single host of 1.1.1.1, the wildcard affinity include an IP address of 1.1.1.1 with a netmask of 255.255.255.255. To specify the range of hosts from 1.1.1.0 to 1.1.1.255, the wildcard affinity would include an IP address of 1.1.1.0 with a netmask of 255.255.255.0, indicating that the first three bytes of the IP address must match exactly and that the last byte is to be ignored.
Wildcard affinity 700 also includes a time to live 722. Time to live 772 is used in the same manner as the time to live for the fixed affinity. Wildcard affinities are deleted by forwarding agents based on the time to live set for the wildcard affinity by the service manager. The timing of such a deletion need not be exact. In one embodiment, the timing need only be accurate to within two seconds. This same tolerance is for fixed affinities as well. Service managers must refresh each wildcard affinity before its time to live expires in order to continue to receive packets that match the wildcard affinity from forwarding agents. As with the fixed affinity, a wildcard affinity may be deleted by sending a duplicate wildcard affinity with a time to live of 0.
Actions
Thus, fixed affinities specify individual flows and packets and wildcard affinities specify sets of flows to be processed in a special way. Such processing is defined by associating actions with the affinities. Actions defined for the affinities specify the service to be performed by the forwarding agent on behalf of the Manager. For fixed affinities, services specified may include:
For Wildcard Affinities, applicable actions are:
Advertise—indicates that the destination IP Address in the Wildcard Affinity is to be advertised by the forwarding agent. This may be done by including the destination IP address in routing protocol updates.
Sequence Number Adjustment—a set of deltas and initial sequence numbers by which the TCP sequence numbers and ACK numbers are to be adjusted.
Forwarding agents may not support all possible actions. For example, some forwarding agents may not support NAT. The set of actions that the service manager expects a forwarding agent to support are identified in an action list which may be included with the wildcard affinity. If the forwarding agent does not support one or more of the actions identified in the list, it discards the wildcard affinity and send a message to the service manager indicating that it does not support all of the actions in the list. This message is referred to as an affinity update deny message. The service manager then may attempt to send a new wildcard affinity that excludes any unsupported actions identified in the affinity update deny message.
Service Messages
Wildcard affinities, fixed affinities, actions, packets, and other messages are sent between service managers and forwarding agents encapsulated in service messages. In one embodiment, messages sent between service managers and forwarding agents are sent using the specific service message format described below. Service messages are sent between service managers and forwarding agents using UDP. Wildcard affinities, which are sent by service managers, can be multicast to a multicast IP Address and UDP Port known to the service manager(s) and forwarding agent(s), or can be unicast to a particular forwarding agent or service manager.
The following service message types are used:
The affinity update-wildcard affinity message is used to send wildcard affinities from a service manager to forwarding agents. The affinity update-fixed affinity message is used to send fixed affinities. The affinity update-deny message is used to report that an affinity update message has been rejected because required actions included in the affinity update are not supported by the receiver. The interest match-wildcard affinity message is used to report a wildcard affinity match to a service manager and the interest match-fixed affinity message is used to report a fixed affinity match to a service manager. The IP packet only message is used to forward an IP packet.
After the service message header, a service message includes one or more message segments. Each message segment begins with its own segment header. FIG. 8B is a diagram illustrating a segment header. Segment header 810 includes a Required flag 812. Required flag 812 defines whether the sender will allow the rest of the message to be processed even if the segment cannot be processed (either because the receiver does not support the function described by the segment or because the receiver does not understand the segment). The required flag either indicates that the segment may be ignored or that the segment is required. If a required segment cannot be processed, then the entire message that includes the segment is dropped and an error message is returned to the sender. Each segment header is followed by data that is specific to the message segment.
The following message segments are used:
The fixed affinity, wildcard affinity and security segments are described immediately below. The remaining segments are described in detail following a description of the message types that include the segments.
Security
If security is expected by the receiver, a security message segment immediately follows the service message header. The security message segment contains the expected security sequence. If the receiver does not expect security, the security message segment is ignored (if present) and the message is accepted. Security is generally not required for IP packet only messages. If authentication is successful the signals are accepted. If the authentication fails, the signal is ignored. Various authentication schemes such as MD5 may be supported. The type of authentication to be used is configured at the senders and receivers, along with a password. If the receiver does not expect authenticated messages, then the security segment may be ignored if it is present and the signal may be accepted whether or not it contains a security segment.
Detailed Message Descriptions
Wildcard Affinity Update
An affinity service precedence field 908 is optionally used to specify the precedence of the service being provided. This allows multiple service managers or a single service manager to send wildcard affinities for different services. An affinity backup precedence field 909 is also optionally used to specify the backup precedence of the service manager that sent the affinity. This allows a backup service manager to send wildcard affinities that are ignored until a higher backup service precedence wildcard affinity that corresponds to a primary service manager is deleted. An identity information segment 910 is optionally used to identify the manager. This information may be used, for example, in an error message on the console of the forwarding agent to indicate which service manager had a problem. A service manager interest data segment is optionally used to include data that should be returned to the service manager when an interest match-wildcard affinity message is sent to the service manager as a result of a forwarding agent determining a wildcard affinity match. Finally, one or more action segments are optionally included. The action segments specify actions that are performed on the packets for the purpose of providing a network service. It should be noted that in some embodiments, fields which are described above as optional may become required and required fields may be optional. This is also generally true of the other message descriptions contained herein.
Fixed Affinity Update
Fixed affinity update message 920 includes a service message header 922. An optional security segment 924 is included as dictated by the needs of the receiver. A fixed affinity segment 926 includes the fixed affinity being sent. An affinity service precedence 928 optionally specifies a service precedence. An affinity backup precedence field 929 is also optionally used to specify the backup precedence of the service manager that sent the affinity. This allows a backup service manager to send affinities that are ignored until a higher backup service precedence affinity that corresponds to a primary service manager is deleted. One or more action segments 930 are optionally included to specify actions to be performed by the receiver for matching packets. An identity information segment 932 is optionally used to identify the service manager that sent the fixed affinity. A service manager interest data segment 934 is optionally used to include data that should be returned to the service manager when an interest match-wildcard affinity message is sent to the service manager as a result of a forwarding agent determining a wildcard affinity match. A forwarding agent interest data segment 936 is optionally used to include data that a forwarding agent requested to be returned to it along with a fixed affinity. Finally, an IP packet segment 938 includes an IP packet.
Usually, the IP packet segment is an IP packet that was sent to a service manager as a result of a wildcard affinity match and that is being sent back to a forwarding agent along with actions to be performed for the packet. In many implementations, the forwarding agent does not devote resources to storing packets that have matched a wildcard affinity and have been forwarded to a service manager. Therefore, the forwarding agent sends the packet to the service manager along with an interest match message and the service manager sends the packet back to the forwarding agent with a fixed affinity update. Thus, the service manager stores the packet for the forwarding agent and returns it to the forwarding agent when the forwarding agent needs to execute an action on the packet. This eliminates the need for storage and garbage collection at the forwarding agent for packets that matched a wildcard affinity and are awaiting instructions from a service manager for handling. In some implementations, the forwarding agents may temporarily store packets that have matched a wildcard affinity. However, it has been found that sending packets to the service manager and having the service manager return packets with fixed affinities simplifies and improves the performance of the forwarding agent.
Affinity Update-deny
Affinity update-deny message 940 includes a service message header 941. An optional security segment 942 is included as dictated by the needs of the receiver. An action list segment 944 includes actions that are not supported by the forwarding agent and that caused the forwarding agent to sent the affinity update-deny message. An affinity segment 946 from the original affinity update that prompted the affinity update-deny message is optionally included. An identity information segment 948 is from the original affinity update that prompted the affinity update-deny message is also optionally included. A service manager interest data segment 950 is optionally used to include data that the service manager sent to the forwarding agent for the forwarding agent to send back to the service manager when an interest match-wildcard affinity message is sent to the service manager. The service manager interest data is used by the service manager to help process the message. A forwarding agent interest data segment 952 is optionally used to include data that the forwarding agent requests to be returned to it along with a fixed affinity.
Interest Match (Wildcard Affinity or Fixed Affinity)
An identity information segment 968 is optionally included from the original affinity update that prompted the interest match message to be sent. A service manager interest data segment 970 is optionally used to include data that the service manager requested when an interest match message is sent to the service manager. A forwarding agent interest data segment 972 is optionally used to include data that a forwarding agent requested to be returned to it along with a fixed affinity. Finally, an IP packet segment is optionally included so that the forwarding agent can send the IP packet that caused the affinity match to the service manager. The IP packet is sent if the corresponding data flag in the interest criteria indicated that the IP Packet should be sent. The IP packet may be sent as a segment of the interest match message or may be forwarded independently in a subsequent IP Packet message, depending on the capabilities of the forwarding agent.
IP Packet Only
IP packet only message 980 includes a service message header 982. An IP Packet segment 984 includes the IP packet. Preferably IP packet only message 980 does not include a security segment, since the flow is essentially just another IP hop and faster forwarding can be achieved without a security segment.
The messages sent between forwarding agents and service managers have now been described in some detail. The wildcard affinity segment, the fixed affinity segment, and the security segment have also been described. The remaining message segments are described in greater detail below in connection with
Thus, the affinity service precedence allows multiple service managers of different types to control the same flow. The value of the precedence dictates the order in which the forwarding agent should process affinities if multiple matches occur. When a matching affinity contains an action that requires the packet to be sent to a service manager, the action is honored. When the packet is returned, the forwarding agent processes the affinity contained in the response and continues with the matching affinity of the next highest precedence.
A sequence delta field 1042 specifies the amount by which the sequence number in packets is to be adjusted. An initial sequence number 1044 specifies the lowest sequence number to which the delta is to be applied. An ACK delta field 1046 specifies the amount by which to adjust the ACK number. An Initial ACK number field 1048 specifies the lowest ACK number to which ACK Delta is to be applied. Thus, sequence numbers and ACK numbers in packets can be modified by forwarding agents according to a scheme determined by a service manager. The scheme is sent to the forwarding agents using the sequence number adjust action segment.
Advertise action segment 1050 includes an advertise address 1052, which is the address to be advertised by the forwarding agent. A subnet mask 1054 may also be used for such advertising. If a subnet mask is used, then the IP address and mask combination indicates a subnet to be advertised. The advertise segment can also be used without specifying a subnet mask.
If the protocol is UDP, then the bits are interpreted as follows:
For other protocols, Bit 15 may be set to indicate all packets.
A data flag 1067 uses the same bit code as the interest mask. Whereas the interest mask determines whether the service manager should be forwarded an interest match message, data flag 1067 specifies whether the service manager is to receive a copy of the packet that caused the interest match with the interest match message. If a bit is set, then the forwarding agent is to send the packet as well as the interest match to interest IP address 1062 and interest port 1064. It should be noted that in some embodiments, the forwarding agents may send messages and forward packets to service managers over a different network so that the interest IP address and interest port may not be used or some other method may be used for specifying where interest match messages and packets should be sent to the service manager.
A copy flag 1068 also uses the same bit code as the interest mask. Each bit specifies whether a copy of the matching packet is to be forwarded to the server. If the bit is set for the packet type, the forwarding agent sends a copy of the matching packet and refers to a hold flag 1069 to determine what to do with the original packet. Hold flag 1069 also uses the same bit code as the interest mask. Hold flag 1069 determines whether the forwarding agent forwards the packet to the service manager or, if possible, holds the packet and waits for the service manager to send a fixed affinity that specifies how the packet should be forwarded by the forwarding agent. If the bit is not set for the packet type, then the forwarding agent forwards the packet. If the bit is set, then the forwarding agent holds the packet, if possible. If the packet cannot be held by the forwarding agent for some reason (e.g., lack of storage) then the forwarding agent forwards the packet to the Manager.
A service message protocol for sending messages and packets between service managers and forwarding agents has been defined in
For example, if a service manager is providing load balancing, the service manager first sends a wildcard affinity update message to a forwarding agent specifying a set of clients that the service manager will load balance. The wildcard affinity may also include an action that directs the forwarding agent to advertise a virtual IP address for a virtual machine that includes all of the load balanced servers. When the forwarding agent intercepts a packet that matches the wildcard affinity, then the forwarding agent sends an interest match message to the service manager. The service manager then determines a server to assign the connection (or the server that has already been assigned the connection) and sends a fixed affinity to the forwarding agent that directs the forwarding agent to dispatch the packet to that server or to use NAT to substitute the server's address in the packet. The service manager also may include an interest criteria in a fixed affinity that specifies that future packets for the flow should not be sent to the service manager, but that the service manager should be notified if certain types of packets such as a FIN or a FIN ACK are received. At any point, the service manager may cancel a fixed affinity or a wildcard affinity sent to a forwarding agent by sending a fixed affinity or a wildcard affinity with a time to live of 0.
Thus service managers are able to control affinities and monitor flows using the above defined messages. When a forwarding agent receives a packet, affinities received from service managers are searched first for the one with the highest service precedence. Once a match is determined, the search order defined for that precedence is used to find another identical Affinity with a better service precedence. If multiple affinities exist with the same best service precedence, they are searched for the one with the lowest backup precedence value.
Service managers manage the storage of affinities on forwarding agents using the time to live portion of the affinity segments. The forwarding agents remove affinities at intervals specified by the service manager if they have not already been removed at the request of a manager (via an affinity update message with a time-to-live of zero). No affinity is kept for an interval longer than the interval specified by the time-to-live set by the manager (within a tolerance of +/−2 seconds in one embodiment) so that the manager can reliably assume that the affinities have been cleared at some small time beyond that interval that accounts for any propagation or processing delays. This simplifies the managing of affinities by the service manager across multiple routers. In some cases, a forwarding agent may need to ask for an affinity again if more traffic arrives for that affinity after it has been deleted.
The service manager itself stores affinities long enough to allow forwarding agents sufficient time to delete their own copies. If an affinity is allowed to expire at a service manager, it must be kept by the service manager long enough so that the forwarding agents have deleted their copies first. This avoids mismatches of affinities across routers should a new affinity assignment request be received while a router still has the old affinity.
Service managers also keep affinities long enough after an outbound FIN is detected for a connection so that the final inbound ACK (or in the case of many Windows web browsers, the inbound RST) can be forwarded to the appropriate host. The use of a ‘sticky’ timer at the service manager satisfies this requirement. If a service manager changes an affinity at a time when it is possible that the affinity is still cached by a forwarding agent, the service manager asks the forwarding agents to delete the affinity before sending the updated affinity.
It should be noted that fixed affinities and wildcard affinities do not themselves include actions in the data structures described above. For flexibility, actions are defined separately but are included with fixed affinities or wildcard affinities in an affinity update message. The associated actions are stored along with the fixed affinity or wildcard affinity on service managers and forwarding agents. Whenever a fixed affinity or a wildcard affinity is referred to as being stored on a forwarding agent or a service manager, it should be understood that associated actions may be stored with the affinity, whether or not such actions are explicitly mentioned.
Likewise, other items may be included in a stored affinity data structure. For example, the affinity may include a time to live when it is sent by a service manager. When the affinity is received by a forwarding agent, the forwarding agent may compute an expiration time from the time to live and store the expiration time along with the fixed affinity.
An architecture that includes service managers and forwarding agents for providing network services has been disclosed. A message protocol for sending messages from service managers to forwarding agents and for reporting activity and forwarding packets from forwarding agents to service managers has been disclosed as well.
Next, it will be described how service managers process packets and messages received from forwarding agents and create modified instructions for the forwarding agents in some circumstances. Service managers thus function as a single point of synchronization for all of the forwarding agents and ensure that each flow is handled in a consistent manner regardless of a specific forwarding agent that happens to handle packets for the flow at any given time.
The service manager determines how packets in each flow are to be handled and keeps track of when packets are received for each flow and when connections are terminated. The service manager provides synchronization to forwarding agents without requiring a two-phase command protocol. Wildcard affinities are repeatedly broadcast by the service manager to forwarding agents and when a fixed affinity sent by a service manager is not received by a forwarding agent, a matching wildcard affinity causes packets to be forwarded to the service manager so that the service manager can re-send instructions included with a fixed affinity as needed.
If the system version is successfully verified, then control is transferred to a step 1110 and the service manager checks the message to determine the message type. If the message type is an IP packet only message, then control is transferred to a step 1112 and the IP packet is processed. If the message is a fixed affinity match, then control is transferred to a step 1116 and the instructions associated with the fixed affinity are carried out. It should be noted that an IP packet may be included with any of the messages sent to the service manager by a forwarding agent and that a service manager may modify the packet or dispatch the packet itself, or return the packet to a forwarding agent to be modified or dispatched. If the message type is an affinity update deny, then control is transferred to a step 1114 and the message is processed. If the message type is a wildcard affinity match, then control is transferred to a step 1118 and the wildcard affinity match message is processed. Steps 1112, 1114, 1116, 1118 are further described in
The error code executed in step 1206 varies according to the application and the network service being provided by the service manager. If the service manager is a firewall, then the service manager will likely drop any IP packet included with the fixed affinity match message or instruct the forwarding agent that sent the fixed affinity match message to drop IP packet. If the service manager is executing load balancing, then it may give the IP packet back to the forwarding agent or direct the forwarding agent to simply forward the packet without modifying the packet. In different applications, the service manager may or may not send a message to the forwarding agent that sent the fixed affinity match indicating that an error occurred.
If a fixed affinity is not found in step 1202 but a wildcard affinity is found in step 1204, then control is transferred to a step 1210 where the service manager updates a state machine to determine what should be done with the flow that corresponds to the fixed affinity match message. The service manager includes a state machine that keeps track of the state of connections and various actions that the service manager has determined should occur for certain flows.
The state machine does different things depending on the service provided by the service manager. For example, if the service manager is providing load balancing, then the state machine keeps track of the various connections being handled by the service manager and the real machines assigned to handle those connections. The service managers that provide load balancing keep track of when connections are closed and when activity occurs on a connection for the purpose making load balancing decisions. A firewall or other security service would keep track of addresses to which connections are allowed, addresses which have been authenticated, or addresses which are being translated.
Once the state machine is updated in step 1210, control is transferred to a step 1212 and the packet is forwarded along with a new fixed affinity update if the fixed affinity corresponding to the packet has been changed. The process then ends at 1214.
If a fixed affinity is found in step 1202, then control is transferred to a step 1220 and the state machine of the service manager is updated. Next, in a step 1222, the packet is forwarded and an affinity update is sent, if necessary, as a result of the updated state machine. The process then ends at 1224.
Thus, when an IP packet only message is received by a service manager, the service manager first attempts to find a fixed affinity that matches the packet. If a fixed affinity is found, then the state machine that the service manager uses to determine how a service is to be provided is updated and the packet is forwarded. If no fixed affinity is found, then the service manager searches for a wildcard affinity and updates the state machine according to whatever wildcard affinity is found. The service manager should not receive IP packet only messages that do not correspond to either a fixed or a wildcard affinity stored on the service manager and if such a message is received, then an error has occurred. A service manager handles an IP packet only messages in a similar manner to how a forwarding agent handles IP packets, with the exception that the service manager includes a state machine that may change or create a fixed affinity in response to packet events.
If the affinity is a wildcard affinity, then control is transferred from step 1106 to a step 1112 and the service manager generates an information message for the administrator. Next, in a step 1114, the service manager determines if the rejected actions may be removed from the wildcard affinity without frustrating the purpose of the service manager. If the actions may be removed, then control is transferred to step 1116 and the service manager sends a new wildcard with revised actions. If the actions may not be removed, then control is transferred to a step 1118 and the process ends. Thus, when an affinity update deny message is received by the service manager, the service manager attempts to send a new wildcard affinity that avoids the rejected actions, if possible.
If the state machine indicates that the fixed affinity should be changed, then control is transferred to step 1414 and the old affinity is deleted and the new affinity is sent in a step 1416 to the forwarding agent. If the state machine in step 1412 does not indicate that the fixed affinity should be changed, then the process ends at 1418.
Thus, each time a fixed affinity is received from a forwarding agent, the service manager first makes sure that the fixed affinity is stored by the service manager and then updates the state machine to determine if the fixed affinity should be changed.
A forwarding agent might return a wildcard affinity interest match for a flow that already has a fixed affinity defined by the service manager if a different forwarding agent had processed prior packets in the flow. The new forwarding agent may not have the fixed affinity that was sent to the previous forwarding agent. The new forwarding agent, however, like all forwarding agents in the multicast group that receives wildcard affinities, does have the wildcard affinity and so the forwarding agent does return a wildcard affinity interest match to the service manager. The service manager finds the correct fixed affinity and updates its state machine and sends the fixed affinity to the new forwarding agent.
Thus, the service manager serves as a point of synchronization for all of the forwarding agents without requiring the forwarding agents to coordinate instructions in a reliable manner using a two phase commit protocol or other method. The service manager receives notification whenever a forwarding agent receives a packet that the service manager has identified as of interest to the service manager and that does not have a corresponding fixed affinity stored on the forwarding agent.
If a fixed affinity is not found, then control is transferred to step 1512 and the forwarding agent looks up the wildcard affinity that corresponds to the wild card interest match. If the service manager does not have a fixed affinity but has a wildcard affinity, then the packet is a new packet for a new flow for which the service manager has not yet determined a fixed affinity using its state machine. The state machine is updated in a step 1514 and an affinity update message is sent in a step 1516, if necessary.
In certain situations, a fixed affinity may not be sent to a forwarding agent for a new flow. For example, if the service manager is acting as a firewall and the connection is to be dropped, then the service manager would simply drop a packet included with the wildcard affinity interest message and might not send a fixed affinity giving specific instructions for dealing with the flow. Future packets in the flow would simply be forwarded to the service manager again as a result of the wildcard affinity interest match and the service manager could continue to discard such packets. Once the affinity update is sent, the process ends at 1518.
If the service manager does not find a wildcard affinity in step 1512, then control is transferred to a step 1520 and an error code is generated. The service manager should not receive wildcard affinity interest matches that do not correspond to wildcards stored on the service manager. The process then ends at 1530.
Thus, the service manager handles wildcard interest match messages by first determining whether a fixed affinity has already been created for the flow. If a fixed affinity exists, then the fixed affinity is updated if necessary and then sent to the forwarding agent which is presumably a new forwarding agent that has not handled previous packets in the flow. The forwarding agent may also be a forwarding agent that has handled previous packets in the flow but that has timed out its fixed affinity. If the packet is a packet that corresponds to a new flow, then the service manager will not have found fixed affinity for that flow. The service manager then looks up the wildcard affinity that caused the interest match and creates a fixed affinity so long as the wildcard interest match corresponds to a wildcard affinity stored on the service manager.
The exemplary state machine process starts at 1602. In a step 1604, the state machine receives a message. Next, in a step 1606 the state machine stores an affinity key that corresponds to the message. Next, in a step 1608, the service manager determines the type of event that the message corresponds to. For example the event could be a fixed affinity match, a wildcard affinity match, a IP packet only message being forwarded, or an affinity update deny message. Next, in a step 1610, the service manager determines a new state based on the content of the message.
The service manager also determines an action based on the new state in a step 1612 and that action is added to a fixed affinity update message, if necessary, in a step 1614. If an old affinity is to be deleted, a multicast of a copy of that affinity with a time to live of zero is sent In a step 1616. Finally, in a step 1618, a new fixed affinity, if one is generated as a result of the state change, is sent to the forwarding agent. The process then ends at 1620.
Thus, the service manager includes a state machine that notes all incoming messages and changes state when appropriate. If the state means that an affinity must be modified, the service manager first deletes the old affinity and then sends an affinity update along with a new action to the forwarding agent.
It should be noted that again, the service manager serves as a single synchronization point for the forwarding agents. The service manager does not need to send a fixed affinity to every forwarding agent when the fixed affinity is assigned. Instead, the service manager relies on the fact that the other forwarding agents will forward packets to the service manager when the packets match a wildcard affinity so long as no fixed affinity is stored by the forwarding agent. The service manager can ensure that forwarding agents will forward packets back to it and essentially check in to receive an updated fixed affinity by simply deleting the fixed affinities from the forwarding agents. Fixed affinities may either be deleted individually or they may be deleted by sending a deletion message for the wildcard affinity that corresponds to the fixed affinity. The deletion of a wildcard affinity causes a fixed affinity to be deleted as well.
Old fixed affinities may be deleted using a copy of the fixed affinity with a time to live of zero broadcast to all forwarding agents. Thus, the service manager does not need to keep track of all of the forwarding agents that have received a specific fixed affinity. The broadcast fixed affinity with a time to live of zero does not interfere with the other forwarding agents. Fixed affinities may also be automatically deleted by forwarding agents when their associated wildcard affinity is deleted or when the fixed affinity is aged out.
Although the foregoing invention has been described in some detail for purposes of clarity of understanding, it will be apparent that certain changes and modifications may be practiced within the scope of the appended claims. It should be noted that there are many alternative ways of implementing both the process and apparatus of the present invention. Accordingly, the present embodiments are to be considered as illustrative and not restrictive, and the invention is not to be limited to the details given herein, but may be modified within the scope and equivalents of the appended claims.
This application is a continuation of U.S. application Ser. No. 09/347,048 filed Jul. 2, 1999 by Mark Albert now U.S. Pat. No. 6,606,315, et al. entitled “SYNCHRONIZING SERVICE INSTRUCTIONS AMONG FORWARDING AGENTS USING A SERVICE MANAGER”. This application is related to co-pending U.S. patent application Ser. No. 09/346,634 entitled DISPATCHING PACKETS FROM A FORWARDING AGENT USING TAG SWITCHING filed concurrently herewith, which is incorporated herein by reference for all purposes; and co-pending U.S. patent application Ser. No. 09/347,124 entitled CASCADING MULTIPLE SERVICES ON A FORWARDING AGENT filed concurrently herewith, which is incorporated herein by reference for all purposes; and copending U.S. patent application Ser. No. 09/347,111 entitled LOAD BALANCING USING DISTRIBUTED FORWARDING AGENTS WITH APPLICATION BASED FEEDBACK FOR DIFFERENT VIRTUAL MACHINES filed concurrently herewith, which is incorporated herein by reference for all purposes; and co-pending U.S. patent application Ser. No. 09/347,428 entitled GATHERING NETWORK STATISTICS IN A DISTRIBUTED NETWORK SERVICE ENVIRONMENT filed concurrently herewith which is incorporated herein by reference for all purposes; and co-pending U.S. patent application Ser. No. 09/347,122 entitled HANDLING PACKET FRAGMENTS IN A DISTRIBUTED NETWORK SERVICE ENVIRONMENT filed concurrently herewith, which is incorporated herein by reference for all purposes; and co-pending U.S. Patent application Ser. No. 09/347,108 entitled SENDING INSTRUCTIONS FROM A SERVICE MANAGER TO FORWARDING AGENTS ON A NEED TO KNOW BASIS filed concurrently herewith, which is incorporated herein by reference for all purposes; and copending U.S. patent application Ser. No. 09/347,126 entitled DISTRIBUTION OF NETWORK SERVICES AMONG MULTIPLE SERVICE MANAGERS WITHOUT CLIENT INVOLVEMENT filed concurrently herewith, which is incorporated herein by reference for all purposes; and co-pending U.S. patent application Ser. No. 09/347,034 entitled INTEGRATING SERVICE MANAGERS INTO A ROUTING INFRASTRUCTURE USING FORWARDING AGENTS filed concurrently herewith, which is incorporated herein by reference for all purposes, and co-pending U.S. patent application Ser. No. 09/347,125 entitled BACKUP SERVICE MANAGERS FOR PROVIDING RELIABLE NETWORK SERVICES IN A DISTRIBUTED ENVIRONMENT filed concurrently herewith, which is incorporated herein by reference for all purpose; and co-pending U.S. patent application Ser. No. 09/347,123 entitled STATEFUL FAILOVER OF SERVICE MANAGERS filed concurrently herewith, which is incorporated herein by reference for all purposes; and co-pending U.S. patent application Ser. No. 09/347,109 entitled NETWORK ADDRESS TRANSLATION USING A FORWARDING AGENT filed concurrently herewith, which is incorporated herein by reference for all purposes, and co-pending U.S. patent application Ser. No. 09/347,036 entitled PROXYING AND UNPROXYING A CONNECTION USING A FORWARDING AGENT filed concurrently herewith, which is incorporated herein by reference for all purposes.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4766606 | Bardutz et al. | Aug 1988 | A |
| 5774660 | Brendel et al. | Jun 1998 | A |
| 5951694 | Choquier et al. | Sep 1999 | A |
| 6006264 | Colby et al. | Dec 1999 | A |
| 6128642 | Doraswamy et al. | Oct 2000 | A |
| 6128657 | Okanoya et al. | Oct 2000 | A |
| 6137777 | Vaid et al. | Oct 2000 | A |
| 6185619 | Joffe et al. | Feb 2001 | B1 |
| 6249801 | Zisapel et al. | Jun 2001 | B1 |
| 6263368 | Martin | Jul 2001 | B1 |
| 6327622 | Jindal et al. | Dec 2001 | B1 |
| 6330602 | Law et al. | Dec 2001 | B1 |
| 6539031 | Ngoc et al. | Mar 2003 | B1 |
| 6594268 | Aukia et al. | Jul 2003 | B1 |
| 6671276 | Bakre et al. | Dec 2003 | B1 |
| 6745246 | Erimli et al. | Jun 2004 | B1 |
| 6781991 | Anderlind | Aug 2004 | B1 |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 09347048 | Jul 1999 | US |
| Child | 10623298 | US |
