VPN failure recovery

Abstract

An approach to rapid failover of a communication path between computers that are linked by redundant virtual links in a virtual private network (VPN) features detection of communication link and device failures through an active monitoring approach and re-routing of communication through a redundant link of the VPN when a failure is detected.

Description

BACKGROUND

This invention relates to recovery from device and communication link failures in a virtual private network (VPN).

Virtual private networks are used to provide virtual communication links, called “tunnels,” between different sites that are coupled by a communication network. Typically, the communication network separating the sites is a public network, such as the Internet, which may not provide secure communication. The tunnels are then established to provide secure communication paths between the sites. One approach to establishing secure tunnels is according to the IPSec standard (see IETF RFCs 2401-2411 and 2451).

Network based services are often configured to have more than one gateway to the public network to provide high availability and communication capacity. For example, a service may have two separate gateways that make use of different communication links to different points-of-presence (POPs) for network service providers on the public network. In this way, if one of the communication links fails, then traffic may be able to pass over the other link. By connecting the gateways to different POPs, this approach also provides redundancy in case a POP fails, or a particular network service provider's network infrastructure (e.g., backbone network) fails. Multiple gateways also provide a basis for balancing communication load, and to reacting to performance degradation, such as congestion, that does not necessarily result in complete failure of a communication path.

Internet Protocol (IP) based networks can make use of dynamic routing protocols to maintain routing information at routing hosts and devices (routers) in the network in order to route IP packets through the network to their destination based on destination address information in packet headers. Two standard protocols for dynamically maintaining routing information are RIP (see RFC 1058) and OSPF (see RFC 2328). One characteristic of these protocols is that if a router determines that a neighboring router is inaccessible, for example because is does not receive responses to a number of requests for routing information from that neighbor, that router updates its routing tables and propagates an indication that the neighbor is inaccessible to its other neighboring routers.

In some virtual private networks, routing information passes between devices over the tunnels such that the tunnels act as virtual links and the devices at each end of the tunnels act as routing devices that are coupled by the virtual links. In this way devices at different sites can dynamically update routing information for paths that use those virtual links.

SUMMARY

One approach to providing fault tolerance in a VPN network is to link computers using redundant virtual links. When communication through one of the virtual links is interrupted, active communication sessions are re-routed to one of the redundant virtual links. In an IP-based virtual network, this re-routing can be performed by passing routing information through the virtual links of the VPN according to the OSPF, RIP, or other dynamic routing protocols. As the routing information is updated, IP packets follow the new route through the redundant virtual link.

Transport layer communication sessions are typically configured to have timeout periods such that if communication is interrupted for more than that timeout period, the transport session is terminated. For example, a TCP session may have a timeout period in the order of 10 seconds. Therefore, if routing information is not dynamically updated in the VPN quickly enough, the transport session may not be rerouted to a redundant virtual link before its timeout period and therefore may be terminated. It is desirable to avoid such interruption of the transport layer sessions.

In general, the invention features an approach to rapid failover of a communication path between computers that are linked by redundant virtual links in a virtual private network (VPN). The approach features detection of communication link and device failures through an active monitoring approach and rerouting of communication through a redundant link when a failure is detected.

In a one aspect, in general, the invention features a method for providing remote access to a server system over a data network. Multiple communication paths through the data network are maintained between each of one or more client systems and the server system. Each path is associated with one of a number of gateway devices that are coupled between the data network and the server system. For each of the client systems, the paths between that client system and the server system are associated with different of the gateway devices. Communication is routed between a first of the client systems and the server system over a first of the communication paths passing through a first of the gateway devices. In the event of a communication failure associated with the first gateway device, the communication is re-routed to a second of the communication paths though a second of the gateway devices.

The method can include one or more of the following features:

The communication paths are through geographically distributed of the gateway devices, and these geographically distributed gateway devices are coupled through a trusted data network.

A separate virtual communication link is maintained between each of the client systems and multiple of the gateway devices.

At least some of the client systems include a client computer coupled to an access device. The virtual links are maintained between the access devices and the gateway devices.

A communication tunnel is maintained between the client system and the gateway device. The tunnel can use the PPTP, L2TP, IPSec, or other tunneling approach.

Network layer communication passes through the tunnel in a Point-to-Point Protocol (PPP) session.

Routing communication between the client system and the server system includes transmitting data from the client system over the data network to a first network addressed of the first gateway device.

Re-routing the communication includes transmitting data from the client system over the data network to a network address of the second gateway device, which is different than the address of the first gateway device.

The method further includes monitoring communication associated with the first gateway device.

Communication characteristics between the first gateway device and one or more devices of the server system are monitored.

The monitoring includes polling the devices from the first gateway device and detecting responses at the first gateway device from those devices.

Polling the devices includes transmitting ICMP echo requests.

Communication characteristics between the first gateway device and the data network are monitored.

A device coupled between the first gateway device and the data network is monitored.

Communication characteristics between the first gateway device and a network Point of Presence (POP) of the data network are monitored.

Communication characteristics between the first gateway device and one or more devices accessible over the data network are monitored.

Communication characteristics between the gateway device and the client system are monitored.

Heartbeat messages from the first gateway device are transmitted to devices and responses from the devices are detected at the first gateway device.

Heartbeat messages include ICMP echo requests, or LCP echo requests.

Re-routing the communication includes terminating the communication paths passing through the first gateway device.

Re-routing the communication includes updating routing data to indicate the second gateway device provides a path between the first client system and the server system.

Updating the routing data includes passing routing data from the second gateway device to the first client system.

Updating the routing data includes passing routing data from the second gateway device to one or more devices, such as routers and host computers, of the server system.

Aspects of the invention can include one or more of the following advantages:

Failover to a redundant link can be rapid enough to avoid timeout of transport layer communication sessions between the computers. For example, a TCP/IP communication session may have a timeout of less than 10 seconds, and the invention provides a mechanism by which communication is reestablished within this timeout period. In this way, ongoing communication, such as a lengthy TCP-based data transfer session does not have to be restarted.

Active monitoring of particularly important communication paths allows IP routing information to be updated quickly in the event of a failure and this updating may be achieved more quickly than may be possible using standard dynamic routing protocols.

Active monitoring using small heartbeat messages may load the communication links less than using standard “hello” messages for dynamic routing update protocols.

Redundant gateway devices do not have to be co-located or tightly coordinated to protect against failures of the gateway devices. For example, for increased reliability, the gateway devices can be geographically separated while still allowing a rapid failover in case of failure of one of the gateway devices or a failure of another device or communication link on the path through the gateway device.

The approach can be combined with redundant device and link backup approaches that allow a backup device or link to take the role of a primary device or link on a failure. If both a primary and backup of such a device fail, the invention provides a mechanism for further failure protection without interrupting communication sessions.

The approach can be combined with control of a variety of types of devices, such as load balancing devices, to route communication through backup gateways.

Other features and advantages of the invention are apparent from the following description, and from the claims.

DESCRIPTION OF DRAWINGS

FIG. 1 is a network diagram that illustrates a number of client computers coupled to a number of server computers using a virtual private network and redundant gateways;

FIG. 2 is a diagram that illustrates possible points of failure in a redundant VPN arrangement;

FIG. 3 is a logical block diagram of a gateway;

FIG. 4 is a diagram that illustrates the use of heartbeat messages in a network;

FIGS. 5A-5C are diagrams that illustrate sequences of events when a device or communication link fails;

FIG. 5A is a diagram that illustrates a gateway detecting a heartbeat failure to an access device;

FIG. 5B is a diagram that illustrates a gateway detecting a heartbeat failure to a server;

FIG. 5C is a diagram that illustrates an access device detecting a heartbeat failure to a gateway;

FIG. 6 is a diagram that illustrates geographically distributed gateways;

FIG. 7 is a diagram that illustrates use of gateways in a redundant server system;

FIG. 8 is a diagram that illustrates use of fault-tolerant gateways.

DESCRIPTION

Referring to FIG. 1, a number of clients 110 communicate with servers 160 over a public network 130, such as the Internet. Clients 110 and servers 160 are typically general-purpose computers. Only two clients 110 are illustrated in FIG. 1; in general, the clients 110 can number in the hundreds or more. Also, three servers 160 are illustrated in FIG. 1; in general, fewer or a greater number of servers are used in various examples of this approach. Public network 130 is a packet network that uses the Internet Protocol (IP) for routing packets through the network. Communication between clients 110 and servers 160 also passes over private network 150. Private network 150 is also a packet network that in this embodiment also uses IP for routing packets. In various examples of this approach, private network 150 may be a single Ethernet LAN or may include a routed network that includes a number of routing hosts or devices on the paths from clients 110 to servers 160. Such a routed network may be geographically distributed.

Each client 110 is configured to communicate through two or more gateways 140, each of which links public network 130 and private network 150. Two representative gateways 140 are shown in FIG. 1. Each gateway 140 has a separate communication link 142 to public network 130 to provide redundancy. For example, each gateway 140 may have a separate dedicated communication link, such as a leased T1 link, that terminates at a separate point-of-presence (POP) of public network 130. Each gateway 140 also has a separate communication link 144 that couples the gateway to private network 150.

Communication between clients 110 and servers 160 passes through public network 130 over virtual links of a virtual private network (VPN). At the site of each client 110, an access device 120 is coupled on the communication path between client 110 and public network 130. Each access device 120 establishes VPN tunnels 135 to two or more gateways 140. In this embodiment, each tunnel makes use of the IPSec protocol to transport and encrypt packets passing between access devices 120 and gateways 140. Each IPSec tunnel carries a PPP (Point-to-Point Protocol) data stream that is transported using a GRE (Generic Routing and Encapsulation) protocol. In this way, each tunnel provides a virtual link between the access device 120 and the gateway 140 at its endpoints. In alternative embodiments, different tunneling protocols, such as native IPSec tunneling, or L2TP or PPTP based tunnels are used to link the access devices and the gateways, and combinations of different tunneling protocols are used to link different clients to the servers.

Each access device 120 is preconfigured with the addresses on public network 130 of two or more gateways 140. For example, the IP host addresses of the gateways are stored in non-volatile memory of the access device. Access device 120 establishes tunnels 135 to those preconfigured gateways when it initially begins to establish communication with servers 160. Access device 120 treats one gateway is treated as a primary gateway while it treats the others as backups. After each tunnel 135 is established, gateway 140 sends updated routing information over private network 150 that identifies that gateway as a providing a route to the client 110 that is accessible over the tunnel. In this embodiment, the routing protocol used by gateways 140 on private network 150 is configurable from a set of protocols that include the OSPF (Open Shortest Path First) protocol and RIP (Router Interchange Protocol).

Client 110 uses its access device 120 as its default IP gateway. IP packets therefore first pass from client 110 to its access device 120. Based on that access device's current primary gateway 140, the access device routes packets addressed to server systems 160 over the tunnel 135 corresponding to its primary gateway 140. The receiving gateway 140 then routes the packet to the destination server 160. IP communication from client 110 to server 160 may use a transport protocol that acknowledges packets, such as TCP, and therefore server 160 sends acknowledgement packets addressed to client 110. These acknowledgments are routed back to client 110 according to routing information at the originating server 160 and at routing devices in private network 150.

Referring to FIG. 2, failures can occur at several points along the communication paths between clients 110 and servers 160. First, the communication path between a primary gateway 140A and a server 160A may fail (292). For example, a communication link or a device such as a router or hub in private network 150 may fail or become overly congested. Note that depending on the redundancy built into private network 150, a communication path from a backup gateway 140B to server 160A may still be available. Another point of failure is primary gateway 140A itself (294). Yet another point of failure is tunnel 135 linking primary gateway 140A and access device 120A (296). For example, the communication link between gateway 140A and public network 130 may fail, the POP that couples that communication link to the public network may fail, or a portion of public network 130 that is associated with that POP (e.g., a network service provider's network backbone) may fail.

The general approach to handling each of these types of failures is to rapidly reroute communication from the tunnel 135 linking access device 120A and primary gateway 140A to the tunnel 135 linking access device 120A and backup gateway 140B. For example, if client 110A had active transport layer sessions with both server 160A and 160B (illustrated as paths 210), then after the failover, these transport sessions follow paths through backup gateway 140B (illustrated as paths 220).

Referring to FIG. 3, gateway 140 includes a number of logical components. In various embodiments, these components are implemented in software, in hardware, of using a combination of hardware and software. An internal interface 310 provides a communication path to private network 150 while an external interface 350 provides a communication path to public network 130. A tunnel module 340 serves as the endpoint for tunnels 135, each passing through external interface 350 to a different access device 120. Each tunnel 135 is associated with a different IP address on the virtual private network. A routing module 320 routes IP packets between private network 150 and tunnels 135. Routing module 320 passes and receives routing information over tunnels 135 using a routing protocol such as RIP. In this way, after a tunnel 135 is set up, routing module 320 receives routing information over the tunnel that identifies the IP address of client 110 that is accessible over that tunnel. When routing module 320 receives an IP packet addressed to a particular client it passes the packet to the appropriate tunnel 135. Routing module 320 propagates routing information it receives from tunnels 135 to private network 150 and routing information it receives from private network 150 to tunnels 135. In this way, devices on private network 150, such as server 160, can make use of the gateway 140 to reach particular clients 110.

Gateway 140 also includes a heartbeat module 330 that communicates with devices on private network 150 and with access devices 120 over tunnels 135. Heartbeat module 330 is configurable. Configuration data 335 includes an identification of the particular devices on private network 150 with which heartbeat module should attempt to maintain communication. In addition, configuration data 335 identifies with which of these devices an ability to communicate is critical. In general, heartbeat module 330 sends periodic heartbeat messages to the identified devices. The period of the heartbeat messages is also configurable, typically in the range of 2.0 to 4.0 seconds. The period is chosen to be long enough to not unduly load the network and is chosen short enough such that a failure can be detected and acted upon quickly. During normal operation, the devices that receive the heartbeat messages immediately respond to each heartbeat message. Configuration data 335 also identifies how many sequential heartbeat messages to a particular device must be unanswered to declare a failure of communication with that device. Typically, this number is in the range of 2-4 messages. Therefore, a period of 3.5 seconds and declaration of a failure after 2 unanswered heartbeats yields detection within approximately 7.0 seconds of a failure, at which point failover is initiated.

Configuration data 335 also identifies whether devices on private network 150 are “critical.” If any critical device fails to respond to a heartbeat message from gateway 140, the entire gateway is disabled and all tunnels 135 from that gateway are shut down and no new tunnels are accepted until the critical device again starts responding to heartbeat messages. When gateway 140 shuts down its tunnels, routing module 320 sends updated routing information to devices on private network 150, for example using the OSPF protocol, to identify the fact that it is no longer a valid gateway to the access devices 120 and clients 110 that it previously served.

Other devices on private network 150 are not marked as critical, and gateway 140 is shut down only after all the non-critical devices concurrently fail to respond to heartbeat messages. Configuration data 335 also identifies the characteristics of heartbeat messages sent to access devices 120 over tunnels 135. If heartbeats through a particular tunnel 135 fail to be answered the configured number of sequential times, that tunnel is terminated, but the gateway otherwise continues to perform its functions. Routing module 320 sends updated routing information to devices on private network 150 to identify the fact that it is no longer a valid gateway to the access device 120 and client 110 that it previously served through the terminated tunnel.

Each access device 120 has a similar logical arrangement as that shown for gateway 140 in FIG. 3. A heartbeat module in an access device sends heartbeats to the gateways 140 to which it has tunnels. If the primary gateways fails to respond, the access device terminates the tunnel and the router module immediately starts passing packets from client 110 over a tunnel 135 to a backup gateway 135. The tunnel module then begins and continues to try to reestablish a tunnel to the primary gateway. The router module at the access device sends a request to the router module at the backup gateway to send updated routing information rather than waiting to periodic routing updates it would send as part of normal operation of the route update protocol (e.g., RIP).

Referring to FIG. 4, the paths for various heartbeat messages (410, 420) are shown in an example with one client 110 and three servers 160 served by two gateways 140. (Each arrow represents a path of heartbeat messages; replies to the heartbeat messages flow in the reverse directions). Each gateway 140 independently sends periodic heartbeat messages 410 to each server 160. In this embodiment, heartbeat messages 410 that are sent from gateway 140 to servers 160 use the PING protocol. Note that in general, private network 150 is a routed network, and therefore each of these heartbeat messages may pass through one or more devices, such as routers, load managers or other routing devices, on the path between the gateway and the server. Therefore, a failure of any of the devices along the path will prevent the heartbeat from being answered.

Each gateway 140 also sends heartbeat messages 420 to each access device 120 to which it has a tunnel connected. Heartbeat messages 420 do not necessarily use the same protocol as do heartbeat messages 410. The protocol is configurable. For example, LCP (Link Control Protocol) Echo Requests are used to implement heartbeats 420.

Access device 120 can also be configured to send heartbeat messages (440) to each of gateways 140. Access devices 120 can also be configured to send heartbeat messages (430) to particular devices (e.g., servers 160) through each gateway 140, thereby directly monitoring critical communication through each of the gateways.

Referring to FIGS. 5A-5C, different types of failures result in different sequences of operations at gateway 140 and access device 120. FIG. 5A illustrates a failure of a heartbeat to an access device. The sequence of events is as follows:

• • 1. Gateway 140 detects a heartbeat failure to an access device 120.
  • 2. The router module of the gateway triggers a routing update to neighboring routing devices on private network 150, using a configurable routing protocol that is appropriate for private network 150 such as OSPF or RIP, which identifies that access device 120 is no longer accessible. This update propagates according to the routing protocol to intermediate routers in private network 150 and to servers 160.
  • 3. Gateway 140 terminates tunnel 135 to access device 120 and then waits to accept a request from the access device to reestablish the tunnel.

Referring to FIG. 5B, a heartbeat failure to a critical server, or to all the non-critical servers, results in the following sequence of events:

• • 1. Gateway 140 detects the heartbeat failure from the server and informs the access devices 120 coupled to it through tunnels 135 that it will be terminating the tunnels to them.
  • 2. The gateway then terminates all the tunnels to the access devices and propagates route update information to routing devices on private network 150 to indicate that it no longer provides a route to the corresponding access devices and clients.
  • 3. At each access device 120 that has had the tunnel to its primary gateway terminated, the router module begins rerouting traffic to the backup gateway. The access device also request updated routing information from the backup gateway.
  • 4. The backup gateway sends the updated routing information in response to the request from the access device.
  • 5. After any tunnel to an access device is terminated, the access device begins to try to reestablish the tunnel. However, the gateway that terminated the tunnel does not accept the requests to reestablish the tunnels until it is one again successfully receiving replies to the heartbeat messages it is sending to servers 160.

If access device 120 is configured to initiate heartbeat messages 430 or 440, the access device itself initiates shutdown of a tunnel when a heartbeat message is not answered. Referring to FIG. 5C, a heartbeat failure from an access device 120 to a gateway 140 results in the following sequence of events.

• • 1. Access device 120 detects a heartbeat failure.
  • 2. The access device terminates the tunnel if it has not already been terminated (for example by the gateway under the scenario shown in FIG. 5B).
  • 3. 4. and 5. are the same from the access device's point of view as when the gateway initiated the termination of the tunnel.

In various examples of this approach, private network 150 can include a variety of different types of routing “fabrics.” For example, private network 150 can be configured to be statically routed or to use a dynamic routing protocol such as OSPF.

Referring to FIG. 6, gateways 140 can be geographically separated, for example, being located in different cities. At the site of each gateway 140, a local area network 620 provides a link between the gateway and a trusted wide area network (WAN) 610. Together, LANs 620 and trusted WAN 610 form a private network such that if communication is re-routed from gateway 140A to gateway 140B, communication passes over trusted WAN 610 to reach server 160.

Referring to FIG. 7, an application of gateways 140 is in a fault tolerant server system in which routers 710 each provide a path to a redundant server system 750. For example, each server system 750 may be located in a different city. Each of the server systems includes a load balancer 720 and a server 730, which uses a data store 740. Servers 730 cooperate to keep the data stores consistent. A gateway 140 is collocated with each server system and is coupled to router 710 at that system to provide an administration interface to maintain the servers and the data stores. Administrators establish secure tunnel connections from public network 130 through the gateways to administration computers 760, which execute applications that are used to maintain the servers and data stores. In the event of a failure of either of the sites, administrators sessions over tunnels to the gateway at that site are failed over to the gateway at the other site.

Referring to FIG. 8, the type of redundancy described above can be combined with hardware failover to a standby gateway in case a primary gateway fails. In FIG. 8, at gateway 840A is a primary gateway that provides a link between router 810A and private network 150. While gateway 840A is functioning, gateway 840B is in a backup role, for example remaining idle or alternatively mirroring operations of primary gateway 840A. In the event of a failure of primary gateway 840A, backup gateway 840B assumes the primary role, and assumes the external addresses of the failed gateway. By assuming the same addresses, routing information in the private network and at the clients does not have to change. Therefore, 840A and 840B together perform the function of a single gateway 140, but with the higher reliability afforded by using a hardware redundancy. Similarly, gateways 840C and 840D together serve the function of a single gateway 140. In the event of other forms of failure, for example failure of router 810A, communication paths are re-routed from gateway 840A to gateway 840C according to the approach described with reference to gateways 140. Therefore, use of redundant tunnels can be combined with redundant approaches for the design of each of the gateways themselves to provide even greater reliability. In yet another configuration, each gateway 140 (or a pair of redundant gateways 840A-B) is coupled to separate routers that are arranged in a primary-backup arrangement using the VRRP protocol (see RFC 2338).

In alternative embodiments, gateways 140 pass information regarding accessibility to various clients 110 using approaches other than using standard dynamic routing protocols. For example, each gateway 140 may communicate directly with a load management device to indicate with clients are accessible using a special purpose communication protocol.

In other embodiments, or in combination with the approaches described above, devices on private network 150 can poll gateways 140 to determine their status. For example, a load balancing device can poll the gateways to determine which access devices 120 are accessible through the gateways.

In alternative embodiments, the primary gateway 140 advertises to devices on private network 150 a lower routing cost for routing traffic to client 110 than does the backup gateway 140. In this way, traffic flowing from private network 150 to client 110 is preferentially directed through the primary gateway 110. Since the primary gateway 140 has advertised a lower cost to route packets to client 110, acknowledgement packets will typically pass back to client 110 through the primary gateway 140 for that client.

It is to be understood that the foregoing description is intended to illustrate and not to limit the scope of the invention. Other embodiments are within the scope of the following claims.

Claims

1. A method for providing remote access to a server system over a data network comprising: maintaining a plurality of communication paths through the data network between each of one or more client systems and the server system, each path being associated with one of a plurality of gateway devices coupled between the data network and the server system, wherein each path is maintained as an active transport layer session;for each of the client systems, the paths between said client system and the server system being associated with different of the gateway devices;routing communication between a first of the client systems and the server system over a first of the communication paths passing through a first of the gateway devices;monitoring communications associated with the first gateway device; andre-routing the communication between the first client system and the server system to a second of the communication paths through a second of the gateway devices, before a transport layer timeout, in the event of a communication failure associated with the first gateway device.

2. The method of claim 1 wherein maintaining the communication paths includes maintaining paths through geographically distributed of the gateway devices, and coupling said geographically distributed gateway devices through a trusted data network.

3. The method of claim 1 wherein maintaining the communication paths includes maintaining a separate virtual communication link between each of the client systems and multiple of the gateway devices.

4. The method of claim 3 wherein at least some of the client systems include a client computer coupled to an access device, and wherein maintaining a virtual link between said client systems and the gateway devices includes maintaining virtual links between the access devices and the gateway devices.

5. The method of claim 3 wherein maintaining each of the virtual links includes maintaining a communication tunnel between the client system and the gateway device coupled by said link.

6. The method of claim 5 wherein maintaining the communication tunnels includes maintaining a PPTP tunnel.

7. The method of claim 5 wherein maintaining the communication tunnels includes maintaining a L2TP tunnel.

8. The method of claim 5 wherein maintaining the communication tunnels includes maintaining a IPSec tunnel.

9. The method of claim 8 wherein maintaining the IPSec tunnel includes passing network layer communication through said tunnel.

10. The method of claim 9 wherein passing network layer communication through the tunnel includes passing network layer communication in a Point-to-Point Protocol (PPP) session and passing the PPP session through the tunnel.

11. The method of claim 1 wherein routing communication between the client system and the server system includes transmitting data from the client system over the data network to a first network addressed of the first gateway device.

12. The method of claim 11 wherein re-routing the communication includes transmitting data from the client system over the data network to a network address of the second gateway device, said address of the second gateway address being different than the address of the first gateway device.

13. The method of claim 1 wherein monitoring communication associated with the first gateway device includes monitoring communication characteristics between the first gateway device and one or more devices of the server system.

14. The method of claim 13 wherein monitoring communication characteristics between the first gateway device and the one or more devices includes polling said devices from the first gateway device and detecting responses at the first gateway device from said devices.

15. The method of claim 14 wherein polling the devices includes transmitting ICMP echo requests.

16. The method of claim 1 wherein monitoring communication associated with the first gateway device includes monitoring communication characteristics between the first gateway device and the data network.

17. The method of claim 16 wherein monitoring communication characteristics between the first gateway device and the data network includes monitoring a device coupled between the first gateway device and the data network.

18. The method of claim 16 wherein monitoring communication characteristics between the first gateway device and the data network includes monitoring communication characteristics between the first gateway device and a network Point of Presence (POP) of the data network.

19. The method of claim 1wherein monitoring communication associated with the first gateway device includes monitoring communication characteristics between the first gateway device and one or more devices accessible over the data network.

20. The method of claim 19 wherein monitoring communication characteristics between the first gateway device and one or more devices accessible over the data network includes monitoring communication characteristics between the gateway device and the client system.

21. The method of claim 19 wherein monitoring communication characteristics between the first gateway device and the one or more devices includes transmitting heartbeat messages from the first gateway device to said devices and detecting responses at the first gateway device from said devices.

22. The method of claim 21 wherein transmitting heartbeat messages includes transmitting ICMP echo requests.

23. The method of claim 21 wherein transmitting heartbeat messages includes transmitting LCP echo requests.

24. The method of claim 1 wherein re-routing the communication includes terminating the communication paths passing through the first gateway device.

25. The method of claim 1 wherein re-routing the communication includes updating routing data to indicate the second gateway device provides a path between the first client system and the server system.

26. The method of claim 25 wherein updating the routing data includes passing routing data from the second gateway device to the first client system.

27. The method of claim 25 wherein updating the routing data includes passing routing data from the second gateway device to one or more devices of the server system.

28. The method of claim 27 wherein passing the routing data to the one or more devices of the server system includes passing said data to routers of said system.

29. The method of claim 27 wherein passing the routing data to the one or more devices of the server system includes passing said data to host computers of said system.

30. A communication system comprising: a plurality of gateway devices, each programmed to maintain a communication path between each of one or more client systems and the server system;the gateway devices being programmed to route communication between a first of the client systems and the server system over a first of the communication paths passing through a first of the gateway devices, communication associated with the first gateway device being monitored, and to re-route the communication between the first client system and the server to a second of the communication paths via an active transport layer session through a second of the gateway devices, before a transport layer timeout, in the event of a communication failure associated with the first gateway device.