Unicast forwarding of adaptive-routing notifications

Abstract

A method for communication includes, in a first network switch that is part of a communication network having a topology, detecting a compromised ability to forward a flow of packets originating from a source endpoint to a destination endpoint. In response to detecting the compromised ability, the first network switch identifies, based on the topology, a second network switch that lies on a current route of the flow, and also lies on one or more alternative routes from the source endpoint to the destination endpoint that do not traverse the first network switch. A notification, which is addressed individually to the second network switch and requests the second network switch to reroute the flow, is sent from the first network switch.

Description

FIELD OF THE INVENTION

The present invention relates generally to communication networks, and particularly to methods and systems for forwarding of adaptive-routing notifications.

BACKGROUND OF THE INVENTION

Various techniques for routing packets through communication networks are known in the art. Some known techniques select routing paths for packets based on the network state, e.g., traffic load or congestion. Such techniques are sometimes referred to as Adaptive Routing (AR). For example, U.S. Pat. No. 8,576,715, whose disclosure is incorporated herein by reference, describes a method for communication that includes routing a first packet, which belongs to a given packet flow, over a first routing path through a communication network. A second packet, which follows the first packet in the given packet flow, is routed using a time-bounded Adaptive Routing (AR) mode, by evaluating a time gap between the first and second packets, routing the second packet over the first routing path if the time gap does not exceed a predefined threshold, and, if the time gap exceeds the predefined threshold, selecting a second routing path through the communication network that is potentially different from the first routing path, and routing the second packet over the second routing path.

U.S. Patent Application Publication 2015/0372916, whose disclosure is incorporated herein by reference, describes a network element that includes circuitry and one or more interfaces. The interfaces are configured to connect to a communication network. The circuitry is configured to assign multiple egress interfaces corresponding to respective different paths via the communication network for routing packets to a given destination-address group, to hold, for the given destination-address group, respective state information for each of multiple sets of hash results, to receive via an ingress interface a packet destined to the given destination-address group, to calculate a given hash result for the packet and identify a given set of hash results in which the given hash result falls, and to forward the packet via one of the multiple egress interfaces in accordance with the state information corresponding to the given destination-address group and the given set of hash results.

U.S. Pat. No. 9,014,006 and U.S. Patent Application Publication 2015/0195204, whose disclosures are incorporated herein by reference, describe a method including receiving in a network switch of a communication network communication traffic that originates from a source node and arrives over a route through the communication network traversing one or more preceding network switches, for forwarding to a destination node. In response to detecting in the network switch a compromised ability to forward the communication traffic to the destination node, a notification is sent to the preceding network switches. The notification is to be consumed by the preceding network switches and requests the preceding network switches to modify the route so as not to traverse the network switch.

SUMMARY OF THE INVENTION

An embodiment that is described herein provides a method for communication including, in a first network switch that is part of a communication network having a topology, detecting a compromised ability to forward a flow of packets originating from a source endpoint to a destination endpoint. In response to detecting the compromised ability, the first network switch identifies, based on the topology, a second network switch that lies on a current route of the flow, and also lies on one or more alternative routes from the source endpoint to the destination endpoint that do not traverse the first network switch. A notification, which is addressed individually to the second network switch and requests the second network switch to reroute the flow, is sent from the first network switch.

In some embodiments, the method further includes receiving the notification by the second network switch, and, in response to the notification, rerouting the flow to one of the alternative routes. In an embodiment, sending the notification includes routing the notification over a route that differs from a reverse of the current route of the flow.

In some embodiments, the topology is a Fat-Tree (FT) topology, in which network switches are arranged in multiple levels including at least a leaf level and a spine level, and in which each route initially traverses an upwards segment that begins at the leaf level and traverses increasing levels, and then traverses a downwards segment that traverses decreasing levels and ends at the leaf level. In an example embodiment, the first network switch belongs to a given level of the FT topology, and identifying the second network switch includes selecting, in an intermediate level that is lower than the given level, an only network switch that lies on the upwards segment of the current route of the flow. In a disclosed embodiment, the intermediate level is one level lower than the given level.

In some embodiments, identifying the second network switch includes holding in the first network switch a data structure that records, per endpoint, a respective network switch in the intermediate level that lies on the upwards segment of a route from that endpoint to the first network switch, and querying the data structure for the network switch associated with the source endpoint. In an embodiment, the data structure is also used for routing packets from the first network switch to destination endpoints. In other embodiments, identifying the second network switch and sending the notification include identifying two or more second switches, and sending respective unicast notifications to the identified two or more second switches.

There is additionally provided, in accordance with an embodiment of the present invention, a network switch in a communication network having a topology. The network switch include multiple ports configured to exchange packets with the communication network, and packet processing circuitry. The packet processing circuitry is configured to detect a compromised ability to forward via the ports a flow of packets originating from a source endpoint to a destination endpoint, to identify, in response to detecting the compromised ability, based on the topology, a second network switch that lies on a current route of the flow, and also lies on one or more alternative routes from the source endpoint to the destination endpoint that do not traverse the network switch, and to send via one of the ports a notification, which is addressed individually to the second network switch and requests the second network switch to reroute the flow.

There is further provided, in accordance with an embodiment of the present invention, a computer software product, the product including a tangible non-transitory computer-readable medium in which program instructions are stored, which instructions, when read by a processor in a first network switch that is part of a communication network having a topology, cause the processor to detect a compromised ability to forward a flow of packets originating from a source endpoint to a destination endpoint, to identify, in response to the compromised ability, based on the topology, a second network switch that lies on a current route of the flow, and also lies on one or more alternative routes from the source endpoint to the destination endpoint that do not traverse the first network switch, and to send from the first network switch a notification, which is addressed individually to the second network switch and requests the second network switch to reroute the flow.

The present invention will be more fully understood from the following detailed description of the embodiments thereof, taken together with the drawings in which:

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram that schematically illustrates a Fat-Tree (FT) network, in accordance with an embodiment of the present invention; and

FIG. 2 is a flow chart that schematically illustrates a method for forwarding Adaptive-Routing Notifications (ARN) in a FT network, in accordance with an embodiment of the present invention.

DETAILED DESCRIPTION OF EMBODIMENTS

Overview

Embodiments of the present invention that are described herein provide improved methods and systems for adaptive routing of packets in communication networks. The embodiments described herein refer mainly to multi-level full FT networks. The disclosed techniques, however, are also applicable in other suitable network topologies such as quasi-FT networks, networks that use Multi-chassis Link Aggregation (MLAG), and multi-port hosts.

In some embodiments, a FT network comprises multiple interconnected network switches that are arranged in levels. The endpoints, i.e., the hosts served by the network, are connected to the switches in the lowest level, also referred to as leaf switches. The switches in the highest level are referred to as spine switches.

Any route through the FT network comprises an “upwards” segment followed by a “downwards” segment. The upwards segment begins at the leaf switch that serves the source endpoint, and proceeds upwards in the order of levels. The downwards segment proceeds downwards in the order of levels, until reaching the leaf switch that serves the destination endpoint.

As will be shown and demonstrated below, full FT networks have the following properties:

• • When routing packets upwards, a switch always has a choice of two or more possible routes en-route to a given destination endpoint. In the downwards direction, on the other hand, a given destination endpoint is reachable from a given switch over only a single possible route.
  • In a given switch in level X, the source address of a packet received from a higher-level switch uniquely defines the switch in level X−1 that forwarded that packet upwards.

The embodiments described herein provide a high-performance adaptive routing scheme that exploits these properties. The description that follows refers to full FT networks simply as FT networks, for the sake of clarity. Generalization to Quasi-FT and other network topologies is addressed further below.

In some embodiments, a current route is set-up for forwarding a flow of packets from a source endpoint to a destination endpoint. At some point in time, a switch along the downwards segment of the current route identifies congestion on the output port used for forwarding the flow, and is therefore compromised in its ability to continue forwarding the packets of the flow over the current route.

Since the congested switch is part of the downwards segment of a full FT network, it cannot reroute the flow locally via a different port. Instead, the congested switch identifies an alternative switch that will reroute the flow. The identified switch (referred to as the “rerouting switch”) belongs to the next-lower level of the FT network and is part of the upwards segment of the current route. In accordance with the FT properties above, this choice guarantees that the rerouting switch has at least one alternative routing option. Moreover, for the particular congested switch, the identity of the rerouting switch is defined uniquely by the identity of the source endpoint (and thus by the source address specified in the packets of the flow).

Typically, the congested switch holds a database that specifies a respective rerouting switch per source endpoint. The congested switch identifies the rerouting switch by querying the database with the source address extracted from the packets of the flow. In some embodiments the same database already exists in the congested switch, for routing packets in the opposite direction. In such embodiments, the database does not need to be created and maintained for the purpose of adaptive routing. An additional attribute may be added, per source address, specifying the address of the rerouting switch to be selected.

Having identified the rerouting switch, the congested switch generates and sends an Adaptive Routing Notification (ARN) that requests the rerouting switch to reroute the flow. The ARN typically comprises a unicast packet that is addressed individually to the rerouting switch. In response to receiving the ARN, the rerouting switch reroutes the flow to an alternative route that reaches the destination endpoint but does not traverse the congested switch.

Unlike other possible solutions, the techniques described herein do not involve sending a notification hop-by-hop in the reverse direction of the current route, or any multicast notification, in an attempt to find a suitable rerouting switch. Instead, in the disclosed embodiments the congested switch uses its knowledge of the network topology to select the appropriate rerouting switch, and then sends a unicast notification that is addressed to that switch. The disclosed techniques are therefore fast and accurate, and incur little traffic overhead.

System Description

FIG. 1 is a block diagram that schematically illustrates a communication network 20, in accordance with an embodiment of the present invention. Network 20 comprises multiple network switches 24 that are interconnected by network links 28. Network 20 provides connectivity and communication services to multiple endpoints 32.

Endpoints 32, also referred to as hosts, may comprise any suitable computing platforms such as servers, workstations or personal computers. Network 20 may operate in accordance with any suitable communication protocol, such as Ethernet or Infiniband.

Switches 24 may comprise network switches, routers or any other suitable network elements that route or forward packets. In the context of the present patent application and in the claims, the terms “switch” and “network switch” refer to any such network element. In most of the embodiments described herein, including the example of FIG. 1, network 20 has a Fat-Tree (FT) topology. In the FT topology, switches 24 are arranged in two or more levels, denoted L0, L1, L2, . . . . Endpoints 32 are connected to switches 24 in the lowest level, denoted L0. This level is referred to as the leaf level. The highest level, in the present example L2, is referred to as the spine level.

An inset at the bottom of FIG. 1 shows the internal structure of switches 24. In this embodiment, each switch comprises multiple ports 36, a switch fabric 40, a control unit 44 and a memory 48. Ports 36 are connected to respective network links 28, and are configured to send and receive packets to and from the network (e.g., to or from an adjacent switch 24 or an endpoint 32). Fabric 40 is configured to forward packets between ports 36. Control unit 44 is configured to control and manage switch 24, including configuring fabric 40 to apply the desired forwarding operations. Memory 48 is configured to store forwarding tables and/or other relevant information.

The configurations of network 20 and switches 24 shown in FIG. 1 are example configurations that are depicted purely for the sake of conceptual clarity. In alternative embodiments, any other suitable network and/or switch configuration can be used. For example, although FIG. 1 shows a three-level FT network, the disclosed techniques can be used in FT networks having any other suitable number of levels. Further alternatively, the disclosed techniques are not limited to FT networks, and can be used in networks having other suitable topologies.

The different elements of switches 24 may be implemented using any suitable hardware, such as in an Application-Specific Integrated Circuit (ASIC) or Field-Programmable Gate Array (FPGA). Memory 48 may comprise, for example, a Random Access Memory (RAM), a Flash memory or other suitable type of memory. In some embodiments, some elements of switches 24 can be implemented using software, or using a combination of hardware and software elements. In the context of the present patent application and in the claims, fabric 40, control unit 44 and memory 48 are referred to as packet processing circuitry that carries out the disclosed techniques. In alternative embodiments, the packet processing circuitry can be implemented in any other suitable manner.

In some embodiments, control units 44 of switches 24 comprise general-purpose processors, which are programmed in software to carry out the functions described herein. The software may be downloaded to the processors in electronic form, over a network, for example, or it may, alternatively or additionally, be provided and/or stored on non-transitory tangible media, such as magnetic, optical, or electronic memory.

Generating Unicast Congestion Notifications

In an FT network, any route from a source endpoint to a destination endpoint comprises an “upwards” segment followed by a “downwards” segment. The upwards segment begins at the switch that serves the source endpoint in the leaf level L0, and proceeds upwards in the order of levels (but not necessarily all the way to the spine level). The downwards segment proceeds downwards in the order of levels, until reaching the switch that serves the destination endpoint in the leaf level L0.

Consider, for example, a route shown in bold in FIG. 1, from a source endpoint 32 denoted S to a destination endpoint 32 denoted D. The upwards segment of this route traverses switches denoted S1, S2 and S3. The downwards segment of this route traverses switches denoted S3, S4 and S5. Another possible route from S to D, marked with dashed lines in the figure, has an upwards segment that traverses switches denoted S1, S6 and S7, and then a downwards segment that traverses switches denoted S7, S8 and S5.

Generally, the full-FT topology has the following properties:

• • Property I: In the upwards direction, switches always have a choice of two or more possible routes en-route to a given destination endpoint. In the downwards direction, on the other hand, a given destination endpoint is reachable from a given switch over only a single possible route.
  • Property II: For a given destination endpoint, any routing decision taken by a switch along the upwards segment unambiguously defines the highest-level switch to be traversed, and the entire downwards segment of the route.
  • Property III: For a given source endpoint, and for a given switch in a level X along the downward segment of a route, the upwards segment can only traverse a single possible switch in any level lower than X. In particular, for a given switch in level X along the downward segment, the upwards segment can only traverse one possible switch in level X−1. Thus, in a given switch in level X, the source address of a packet received from a higher-level switch uniquely defines the switch in level X−1 that forwarded that packet upwards. This property is based on a more fundamental property—A packet forwarded downwards via a specific switch in level X uniquely defines (for a given source address) another switch in level X via which the packet was forwarded upwards. For a specific source address, to pass via a specific switch at level X in the upward segment, there is only a single possible choice of a switch at level X−1 in the upward segment.

In some embodiments of the present invention, switches 24 carry out an adaptive routing scheme that exploits the above properties. Consider a flow of packets that originates from a source endpoint S and is destined to a destination endpoint D. The flow is initially routed through FT network 20 along a certain route having an upwards segment and a downwards segment.

In an example embodiment, a switch 24 that lies on the downwards segment of the route encounters congestion at the output port it uses to forward the flow downwards. The congestion prevents the switch from continuing to forward the packets of the flow over the current route. Since the switch in question is on the downwards segment, it cannot choose an alternative route that reaches the same destination endpoint (see PROPERTY I above).

In order to recover from this situation, the switch selects an alternative switch along the current route, and requests the alternative switch to reroute the flow. In the description that follows, the former switch is referred to as the congested switch, and the latter switch is referred to as the rerouting switch.

Because of PROPERTY I above, the rerouting switch should lie on the upwards segment of the current route, so that it will have at least one alternative routing option. In an embodiment, the congested switch is on level X of the FT network. The congested switch chooses a switch in level X−1 that lies on the upwards segment of the current route, to serve as the rerouting switch. In accordance with PROPERTY III above, only a single switch in level X−1 lies on the upwards segment of the current route, and, for a given congested switch, the identity of this rerouting switch is uniquely defined by the identity of the source endpoint (and thus by the source address of the packets in the flow).

In some embodiments, each switch 24 holds a database or any other suitable data structure that records, per source address, the identity of the switch in the next-lower level that will serve as the rerouting switch. The database is typically stored in memory 48 of the switch. Note that a given source address may be mapped to different rerouting switches in the databases of different switches. In a given switch, however, each source address is mapped to a unique respective rerouting switch.

Each switch 24 may use any suitable technique for constructing the database, i.e., for obtaining a mapping between each source address and a respective rerouting switch in the next-lower level. In one embodiment, this mapping already exists in the switch—It is the same mapping used for forwarding packets in the opposite direction to this endpoint. Alternatively, the database may be pre-programmed into each switch, or learned adaptively during operation. In some embodiments, the addressing scheme used in network 20 is location-based, in which case database 24 may be simplified.

Thus, when a need arises to reroute a flow, the congested switch queries its database with the source address of the flow, and retrieves the identity (e.g., the address) of the rerouting switch. The congested switch then generates a notification packet, referred to as “adaptive routing notification (ARN),” “congestion notification” or simply “notification.” The ARN comprises a unicast packet that is addressed individually to the specific rerouting switch selected by the congested switch.

The congested switch sends the ARN to the rerouting switch. The rerouting switch receives the ARN, and in response may reroute the flow to an alternative route that reaches the destination endpoint but does not traverse the congested switch. Note that, since the ARN is addressed explicitly to the rerouting switch, it can be forwarded to the rerouting switch over any desired route, not necessarily over the reverse direction of the route of the flow.

For example, with reference to FIG. 1, consider a scenario in which switch S4 (on the downwards segment of the current route from S to D marked in bold) is the congested switch. Switch S4 is in level L1, and therefore the rerouting switch is one of the switches in level L0. In accordance with the database in switch S4, the rerouting switch is necessarily switch S1. Switch S4 thus sends an ARN to switch S1. In response, switch S1 reroutes the flow to an alternative route (shown in dashed lines) that reaches endpoint D but does not traverse the congested switch S4.

FIG. 2 is a flow chart that schematically illustrates a method for forwarding ARNs in FT network 20, in accordance with an embodiment of the present invention. The method begins with a switch 24 at level X of FT network 20 receiving from level X+1 packets of a certain flow, at a packet reception step 60. The packets originate from source endpoint S, and are to be forwarded downwards to level X−1 en-route to destination endpoint S.

At a congestion checking step 64, the switch checks for congestion at the egress port designated for forwarding the packets of the flow. If no congestion exists, the method loops back to step 60 above.

If congestion is detected, the congested switch queries its database to identify the appropriate rerouting switch in level X−1, at a rerouting identification step 68. At a notification step 72, the congested switch generates and sends a unicast ARN, which is addressed individually to the identified rerouting switch. Subsequently, the rerouting switch receives the ARN and reroutes the flow.

In the example above, a congested switch in level X of the FT network selects a rerouting switch in level X−1. In alternative embodiments, a congested switch in level X may select a rerouting switch in any level that is lower than X, e.g., in level X−2 or X−3 (if such levels exist). The latter choice of rerouting switch will also result in a route that does not traverse the congested switch, but may also reroute some additional traffic that did not traverse the congested switch in the first place.

The examples above refer mainly to a congested switch in the downstream segment. In alternative embodiments, the disclosed techniques can also be carried out in a switch that is part of the upstream segment, but is nevertheless unable to reroute the traffic locally. For example, in such a switch all possible egress ports leading to the possible alternative routes may be congested. For example, the aggregate bandwidth over the upstream ports of the switch (the ports connecting to upper-level switches) may be smaller than the aggregate bandwidth over the downstream ports (the ports connecting to lower-level switches). This scenario is sometimes referred to as oversubscription. In such an embodiment, being aware of the network topology, the congested switch may select the previous switch in the upwards segment as the rerouting switch, and send a unicast ARN to that switch.

As noted above, the disclosed techniques are not limited to full FT networks. For example, in some embodiments the disclosed techniques are implemented in a quasi-FT network. Unlike full FT, in a quasi-FT network a switch in the downwards segment of a route may have one or more options for rerouting in case of congestion on the current egress port. In such cases, the terms “congestion” or “compromised ability to forward packets” refers to congestion or compromised ability on all possible egress ports. Moreover, in a quasi-FT network, for a given congested switch and a given source address, there may exist two or more rerouting switches. Thus, in some embodiments the congested switch identifies two or more rerouting switches, and sends a respective unicast ARN to each of them.

In some embodiments, source endpoint S is connected to two or more leaf switches in level L0, for example using different ports of a multi-port Network Interface Controller (NIC) of the endpoint. This sort of configuration is sometimes referred to as Multi-chassis Link Aggregation (MLAG). When using MLAG, packets originating from the source endpoint may enter network 20 via two or more different switches, and thus traverse two or more different routes to the destination endpoint D. Therefore, in these embodiments the congested switch identifies two or more rerouting switches (one per each of the two or more current routes from S to D), and sends the ARN to each of the rerouting switches.

Although the embodiments described herein refer mainly to congestion control, the disclosed techniques can also be used for mitigating failures such as port failure or link failure. In the present context, both congestion and failure are referred to herein as “compromised ability” of a switch to forward packets to a destination endpoint. Upon detecting compromised ability to forward packets, a switch may use the disclosed techniques for identifying a rerouting switch and sending a unicast ARN to the rerouting switch.

It will be appreciated that the embodiments described above are cited by way of example, and that the present invention is not limited to what has been particularly shown and described hereinabove. Rather, the scope of the present invention includes both combinations and sub-combinations of the various features described hereinabove, as well as variations and modifications thereof which would occur to persons skilled in the art upon reading the foregoing description and which are not disclosed in the prior art. Documents incorporated by reference in the present patent application are to be considered an integral part of the application except that to the extent any terms are defined in these incorporated documents in a manner that conflicts with the definitions made explicitly or implicitly in the present specification, only the definitions in the present specification should be considered.

Claims

1. A method for communication in a communication network that includes network switches arranged in multiple levels, the method comprising: detecting, in a first network switch, a compromised ability of the first network switch to forward a flow of packets originating from a source endpoint to a destination endpoint over a route of network switches;in response to detecting the compromised ability, selecting, by the first network switch, a second network switch in the route, which is not an immediately previous hop in the route; andsending from the first network switch to the second network switch, a unicast notification addressed only to the second network switch, requesting the second network switch to reroute the flow.

2. The method according to claim 1, and comprising receiving the unicast notification by the second network switch, and, in response to the unicast notification, rerouting the flow to an alternative route.

3. The method according to claim 1, wherein sending the unicast notification comprises routing the unicast notification over a second route that differs from a reverse of the route.

4. The method according to claim 1, wherein the communication network has a Fat-Tree (FT) topology, wherein the multiple levels comprise at least a leaf level and a spine level, and wherein each route initially traverses a respective upwards segment that begins at the leaf level and traverses increasing levels, and then traverses a respective downwards segment that traverses decreasing levels and ends at the leaf level.

5. The method according to claim 1, wherein sending the unicast notification comprises holding in the first network switch a data structure that records, per endpoint, a respective address of a network switch that (i) is in the level that is one level below the first network switch and (ii) lies on the upwards segment of a route from that endpoint to the first network switch, querying the data structure for an address of the network switch associated with the source endpoint, and addressing the unicast notification to the queried address.

6. The method according to claim 5, wherein the data structure is also used for routing packets from the first network switch to destination endpoints.

7. A network switch in a communication network that includes network switches arranged in multiple levels, the network switch comprising: multiple ports, configured to exchange packets with the communication network; andpacket processing circuitry, configured to: detect a compromised ability of the network switch to forward a flow of packets originating from a source endpoint to a destination endpoint over a route of network switches; andin response to detecting the compromised ability, select a second network switch in the route, which is not an immediately previous hop in the route, and send via one of the ports a unicast notification addressed only to the second network switch, requesting the second network switch to reroute the flow.

8. The network switch according to claim 7, wherein the packet processing circuitry is configured to send the unicast notification over a second route that differs from a reverse of the route of the flow.

9. The network switch according to claim 7, wherein the communication network has a Fat-Tree (FT) topology, wherein the multiple levels comprise at least a leaf level and a spine level, and wherein each route initially traverses a respective upwards segment that begins at the leaf level and traverses increasing levels, and then traverses a respective downwards segment that traverses decreasing levels and ends at the leaf level.

10. The network switch according to claim 7, wherein the packet processing circuitry is configured to hold in a data structure that records, per endpoint, a respective address of a network switch that (i) is in the level that is one level below the first network switch and (ii) lies on the upwards segment of a route from that endpoint to the first network switch, to query the data structure for an address of the network switch associated with the source endpoint, and to address the unicast notification to the queried address.

11. The network switch according to claim 10, wherein the packet processing circuitry is configured to use the data structure for routing packets from the network switch to destination endpoints.

12. A computer software product, the product comprising a tangible non-transitory computer-readable medium in which program instructions are stored, which instructions, when read by a processor in a first network switch that is part of a communication network including network switches arranged in multiple levels, cause the processor to: detect a compromised ability of the first network switch to forward a flow of packets originating from a source endpoint to a destination endpoint over a route of network switches; andin response to detecting the compromised ability, select a second network switch in the route, which is not an immediately previous hop in the route, and send a unicast notification addressed only to the second network switch, requesting the second network switch to reroute the flow.

13. The method according to claim 4, wherein the first network switch is located on the downwards segment of the route and wherein selecting the second network switch comprises selecting a network switch on an upwards segment of the route.

14. The method according to claim 4, wherein selecting the second network switch comprises selecting a network switch which belongs to a level of the communication network that is exactly one level below the first network switch.

15. The method according to claim 1, wherein selecting the second network switch comprises querying a database in the first network switch with an address of the source endpoint to receive an identity of the second network switch.

16. The method according to claim 1, wherein selecting the second network switch comprises querying a forwarding table in the first network switch.