Techniques for selecting paths and/or trunk ports for forwarding traffic flows

Abstract

Techniques that offer enhanced diversity in the selection of paths (e.g., ECMP paths) and/or ports from ports associated with trunks for forwarding data traffic. In one embodiment, one or more functions are used to generate a result. A first portion of the generated result may be used as an index (e.g., ECMP index) for selecting a path (e.g., an ECMP path) from multiple possible paths for forwarding a packet. A second portion of the generated result, different from the first portion, may be used as an index (trunk index) for selecting an output port from multiple output ports associated with a trunk for forwarding a packet. In this manner, selected portions of the generated result may be used as indices, one for selecting a path and another for selecting a trunk port for forwarding packets such that the two indices are not the same and are not dependent upon one another.

Description

BACKGROUND OF THE INVENTION

Embodiments of the present application relate to forwarding data packets in a computer network, and more particularly to techniques for selecting a path such as an Equal Cost MultiPath (ECMP) path and/or a trunk port for forwarding data packets.

Network devices commonly have to select a path from multiple available choices for forwarding a packet. For example, ECMP is a routing strategy where next-hop packet forwarding to a destination can occur over multiple equal cost routing paths. The effect of multipath routing on a network device (e.g., a router) configured to forward packets is that the network device potentially has several next-hops for any given destination and must use some method to choose which path to the next-hop should be used for a given data packet. The use of ECMP helps to reduce delay and congestion in networks by taking advantage of multiple paths through a network by splitting traffic flows across those paths. Accordingly, a network device such as a router has to be able to select a particular ECMP path from multiple available paths to forward a packet.

Trunking is another technique that is commonly used in networks. A trunk represents a logical collection of multiple output ports generally associated with the same route or connected to the same MAC address. In a network environment, when a selected output path for a packet is a trunk, a network device has to be able to select a port from multiple ports associated with the trunk for forwarding the packet. In certain network environments, a selected ECMP path may itself correspond to a trunk. In such an environment, in addition to selecting a particular ECMP path, the network device also has to select a particular output port from the multiple output ports associated with the trunk for forwarding the data packet.

Conventionally, selection of paths (e.g., ECMP paths) and/or trunk ports for data forwarding is done by simply hashing on various fields of a packet header, such as based upon the IP source and destination address fields, and using the hash for selecting the path and/or trunk port. The diversification in the selection offered by such conventional techniques however is quite poor and does not offer proper distribution of traffic to the available paths and/or trunk ports. For example in ECMP forwarding, poor diversification results in the same ECMP path and/or trunk port being selected for forwarding the traffic flow packets at multiple stages of the network. As a result, a router forwards traffic with the same source and destination addresses using the same port of a trunk or the same path, not fully utilizing the bandwidth available for the traffic via other ports or paths available to the router. Accordingly, using conventional techniques in which all routers in the network derive their hashing decision purely based on information extracted from the packet header, correlation occurs among routers that any given packet flow traverses, and such correlation reduces diversification. Further, existing ECMP solutions only work in some network topologies and provide limited diversification.

BRIEF SUMMARY OF THE INVENTION

Embodiments of the present invention provide techniques that offer enhanced diversity in the selection of paths (e.g., ECMP paths) and/or ports from ports associated with trunks for forwarding data traffic. In one embodiment, one or more functions are used to generate a result. A first portion of the generated result may be used as an index (path index) for selecting a path (e.g., an ECMP path) from multiple possible paths for forwarding a packet. A second portion of the generated result, different from the first portion, may be used as an index (trunk index) for selecting an output port from multiple output ports associated with a trunk for forwarding a packet. In this manner, selected portions of the generated result may be used as indices, one for selecting a path and another for selecting a trunk port for forwarding packets such that the portions used for the two indices are not the same. The path index and the trunk index are not dependent upon one another.

According to an embodiment of the present invention, techniques are provided for determining a path for forwarding a packet. A result may be generated using one or more sections of the packet and one or more functions. A first portion of the result may then be used to select a path from a plurality of paths for forwarding the packet from the network device, wherein the first portion of the result is less than the whole result. In one embodiment, a plurality of paths may be equal-cost-multipath (ECMP) paths and the first portion is used to select an ECMP path from the plurality of ECMP paths.

The one or more functions that are used to generate a result may comprise different functions such as a cyclic redundancy check (CRC) function, a rotate function, and others. In one embodiment, a CRC function is applied to the one or more sections of the packet to generate the result. In another embodiment, multiple functions may be used to generate the result. For example, a rotate function may be applied to the one or more sections of the packet to create a rotated result and a CRC function may be applied to the rotated result to generate the result. In another embodiment, another rotate function may be applied to the CRC result to generate the result, a portion of which is then used to select a path for forwarding the packet. The amount of rotation applied by a rotate function may be preconfigured for the network device performing the processing. Different network devices may be configured to apply different amounts of rotation.

In one embodiment, a second portion of the result may be used to select a port from a plurality of ports associated with a trunk for forwarding the packet, wherein the second portion is different from the first portion. The first and second portions though different may still have some overlapping sections. In an alternative embodiment, the first portion and second portion are non-overlapping portions of the result.

According to an embodiment of the present invention, techniques are provided for selecting a port for forwarding a packet. A result may be generated using one or more sections of the packet and one or more functions. A portion of the result may be used to select a port from a plurality of ports associated with a trunk for forwarding the packet from a network device, wherein the portion is less than the whole result.

The one or more functions that are used to generate a result may comprise different functions such as a cyclic redundancy check (CRC) function, a rotate function, and others. In one embodiment, a CRC function is applied to the section of the packet to generate the result. In another embodiment, multiple functions may be used to generate the result. For example, a rotate function may be applied to the section of the packet to create a rotated result and a CRC function may be applied to the rotated result to generate the result. In another embodiment, another rotate function may be applied to the CRC result to generate the result, a portion of which is then used to select a path for forwarding the packet. The amount of rotation applied by a rotate function may be preconfigured for the network device performing the processing. Different network devices may be configured to apply different amounts of rotation.

The foregoing, together with other features and embodiments will become more apparent when referring to the following specification, claims, and accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 depicts a simplified block diagram of a network device that may incorporate an embodiment of the present invention;

FIG. 2 depicts a simplified flowchart depicting a method for generating indices used for selecting a path/and or a trunk port from multiple ports associated with a trunk for forwarding a packet according to an embodiment of the present invention;

FIG. 3 depicts a simplified flowchart depicting a method for generating a path index and/or a trunk index using multiple functions according to an embodiment of the present invention;

FIG. 4 depicts a simplified block diagram of modules that may be used to implement an embodiment of the present invention;

FIG. 5 depicts a simplified flowchart depicting a method of using an ECMP index to select an ECMP path according to an embodiment of the present invention;

FIG. 6 depicts a simplified flowchart depicting a method of using a trunk index to select an output port from multiple output ports associated with a trunk according to an embodiment of the present invention; and

FIG. 7 depicts a multistage network that may incorporate an embodiment of the present invention.

DETAILED DESCRIPTION OF THE INVENTION

In the following description, for the purposes of explanation, specific details are set forth in order to provide a thorough understanding of embodiments of the invention. However, it will be apparent that the invention may be practiced without these specific details.

Embodiments of the present invention provide techniques that enhance diversification in the selection of paths and/or trunk ports for forwarding traffic flows comprising packets. FIG. 1 depicts a simplified block diagram of a network device 100 that may incorporate an embodiment of the present invention. In the embodiment depicted in FIG. 1, network device 100 comprises a plurality of ports 102, a media access controller (MAC) 104, a packet processor 106, one or more memories associated with packet processor 106 such as a content-addressable memory (CAM) 110, a parameter RAM (PRAM) 112, a table 114 (referred to as CAM2PRAM), a traffic manager 108, and a management processor 116 with associated memory 118. The components of network device 100 depicted in FIG. 1 are meant for illustrative purposes only and are not intended to limit the scope of the invention in any manner. Alternative embodiments may have more or less components. For example, while only one packet processor 106 is depicted in FIG. 1, alternative embodiments may have multiple packet processors.

Network device 100 receives and transmits data flows comprising packets using ports 102. A port within ports 102 may be classified as an input port or an output port depending upon whether a packet is received or transmitted using the port. A port over which a packet is received by network device 100 is referred to as an input port. A port used for communicating or transmitting a packet from network device 100 is referred to as an output port. A particular port may function both as an input port and an output port. Ports 102 may be capable of receiving and/or transmitting different types of data traffic at different speeds including 1 Gigabit/sec, 10 Gigabits/sec, or more. In some embodiments, multiple ports of network device 100 may be logically grouped into one or more trunks. A trunk represents a logical collection of multiple output ports of a network device generally associated with the same route or connected to the same MAC address.

In one embodiment, network device 100 may receive one or more packets via one or more input ports. For a packet received over an input port, network device 100 may be configured to determine an output port for the packet. The packet may then be forwarded to the determined output port and transmitted from network device 100 using the output port. As part of the processing to determine an output port for a packet, network device 100 may be configured to select a particular path (e.g., an ECMP path) from multiple paths (e.g., multiple ECMP paths) that may be available for forwarding the packet and select an output port corresponding to the particular selected path. If the selected path is a trunk, as part of forwarding a packet from an input port to an output port, network device 100 may be configured to select a particular output port from the multiple output ports associated with the selected trunk for forwarding the packet. The packet may then be communicated from network device 100 using the selected trunk output port.

In the embodiment depicted in FIG. 1, ports 102 are coupled to a media access controller (MAC) 104. Packets received by network device 100 via one or more ports 102 may be forwarded to MAC device 104 and then to packet processor 106 for further processing. MAC 104 provides an interface between ports 102 and packet processor 106.

Packet processor 106 is configured to process each packet received by network device 100 and determine how the packet is to be forwarded. This processing may involve performing lookups in CAM 110, PRAM 112, and CAM2PRAM 114. In one embodiment, as part of the processing, packet processor 106 is configured to determine an output port to which the packet is to be forwarded. As part of determining an output port to which the packet is to be forwarded, packet processor 106 is configured to select a particular path from multiple paths for forwarding the packet and then select an output port corresponding to a particular selected path. For example, packet processor 106 may be configured to select an ECMP path from multiple ECMP paths for forwarding the packet and forward the packet to a port corresponding to the selected ECMP path. If the selected path is a trunk, as part of determining an output port to which the packet is to be forwarded, packet processor 106 is configured to select a particular output port from the multiple output ports associated with the selected trunk for forwarding the packet. The packet may then be communicated from network device 100 using the selected output port. In some scenarios, the selected path may itself be a trunk. In such a scenario, upon selecting a particular path, packet processor 106 is configured to select a particular output port from the multiple output ports associated with the selected path for forwarding the packet.

According to an embodiment of the present invention, packet processor 106 performs the selection of a path and/or selection of a port from ports grouped as a trunk using techniques that provide enhanced diversification in the selection of the path and/or trunk port. Embodiments of the present invention provide enhanced diversification by more equally distributing traffic flows between available paths (e.g., ECMP paths) and/or trunk ports.

In one embodiment, packet processor 106 is configured to extract one or more sections of the packet to be forwarded. Packet processor 106 is then configured to generate a result using the extracted sections of the packet and one or more functions. A portion of the generated result is used as an index (path index) for selecting a path from multiple possible paths for forwarding the packet. For example, a portion of the generated result may be used as a path index to select an ECMP path from multiple ECMP paths. A path index used for selecting an ECMP path may also be referred to as an ECMP index.

Another portion of the generated result may be used as an index (referred to as a trunk index) for selecting an output port from multiple output ports associated with a trunk for forwarding the packet. The portions of the generated result are selected such that the portion used as the path index is not the same as the portion used as the trunk index. It is possible that the portion selected as the path index and the portion selected as the trunk index have some overlap. In some embodiments, the portions are selected such that there is no overlap. As a result of the manner in which the result is generated and portions of the result used as the path index and the trunk index, the two indices are not dependent upon one another. Further details related to generating the results and using portions of the generated result as indices are described below.

As depicted in FIG. 1, network device 100 may comprise a management processor 116 that is configured to perform housekeeping and management functions related to network device 100. For example, programming and maintenance of tables and information in CAM 110, PRAM 112, and CAM2PRAM 114 may be performed using management processor 116. In one embodiment, management processor 116 may be a general purpose microprocessor such as a PowerPC, Intel, AMD, or ARM microprocessor, operating under the control of software stored in a memory 118 accessibly coupled to the management processor 116.

Since processing performed by packet processor 106 needs to be performed at a high packet rate in a deterministic manner, packet processor 106 is generally a dedicated hardware device configured to perform the processing. In one embodiment, packet processor 106 may be a programmable logic device such as a field programmable gate array (FPGA). Packet processor 106 may also be an ASIC.

In one embodiment, network device 100 may comprise multiple linecards, with each linecard comprising the components depicted in FIG. 1. In such an embodiment, an output port to which a packet is forwarded for transmission from the network device may lie on the same linecard as the input port or on a different linecard. Traffic manager 108 is configured to facilitate communications between the different linecards.

FIG. 2 depicts a simplified flowchart 200 depicting a method for generating indices used for selecting a path/and or a trunk port from multiple ports associated with a trunk for forwarding a packet according to an embodiment of the present invention. In one embodiment, the processing depicted in FIG. 2 is performed by packet processor 106 depicted in FIG. 1. In other embodiments, the processing may be performed by one or more components of network device 100, including or not including packet processor 106. The processing may be performed by software executed by a processor, hardware, or combinations thereof.

As depicted in FIG. 2, one or more sections of the packet to be forwarded are extracted (step 202). The extracted sections may include, for example, sections from the header of the packet, sections from the payload section of the packet, and/or combinations of sections selected from different parts of the packet. In one embodiment, 256-bits from the header of the packet are extracted. For a network device, packet processor 106 may be preconfigured to extract specific sections of the packet in 202. The sections of a packet that are extracted in 202 may also depend on the type of packet that is to be forwarded. For example, the sections of a packet extracted for a packet of a first type (e.g., IPv4 packet) may be different from the sections extracted for a packet of a different type (e.g., IPv6 packet). The sections of a packet that are to be extracted may be user-programmable.

A function is then applied to the sections of the packet extracted in 202 to generate a result (step 204). Various different functions may be used in 204. In one embodiment, the function selected is such that the result generated by the function is sensitive to the inputs to the function such that even a small change in the inputs causes a change in the result that is generated. In one embodiment, a cyclic redundancy check (CRC) function is used. A CRC function takes inputs of different lengths and generates an output result that is generally of a certain fixed size. The extracted portions of the packet are provided as input to a CRC function that generates a CRC result based upon the inputs. Due to the nature of a CRC function, the CRC result is very sensitive to the inputs that are provided to the CRC function—any change in the input to the CRC function, even a one bit change, changes the resultant CRC value. Accordingly, if there is even a slight change in the extracted portions of the packet that are provided as input to the CRC function, then the resultant CRC result will be different.

For example, in one embodiment, 256 bits of the packet header may be selected in 202 and provided as input to a CRC function. In one embodiment, a 32-bit CRC function is used that generates a 32-bit CRC result. Various different techniques may be used for generating the 32-bit CRC result. In one embodiment, the polynomial used is (0 1 2 4 5 7 8 10 11 12 16 22 23 26 32)X³²+X²⁶+X²³+X²²+X¹⁶+X¹²+X¹¹+X¹⁰+X⁸+X⁷+X⁵+X⁴+X²+X+1

The following pseudo code may be used for generating a 32-bit CRC result according to an embodiment of the present invention (the input data in this example code is assumed to be 256 bits wide):

crc_tmp[31:0]=0 (initial seed can be programmed for multi-level system purpose)

crc_mask=32′h04C1_—1DB7 (polynomial)

for (i=0; i<256; i++)/*applying to each bit of the 256 bits selected from packet*/

crc_tmp=data[i] ^((crc_tmp & 0x8000_—0000)>>31)?((crc_tmp<<1)^crc_mask):(crc_tmp<<1);

After the running of the algorithm, the crc_tmp stores the generated CRC result.

Besides a CRC function, other functions may also be used in 204. These functions may include a rotate function, a bit swap function, a bit inverting function, a bit XOR function, and the like.

A portion of the result generated in 204 is then used as path index to select a path, from possibly multiple paths, for forwarding the packet to a next-hop device (step 206). As part of 206, an output port corresponding to the selected ECMP path may also be selected. In one embodiment, the portion of the result used in 204 as the path index is less than the full result. For example, if the CRC result generated in 204 is 32-bits wide, then the path index may be less than 32 bits (e.g., bits 0-15 of the generated result may be used as the path index). The portion used in 206 may be a contiguous portion of the result generated in 204 or may be made up of non-contiguous portions of the result. An example of using non-contiguous portions as the path index is using bit 0-4 and 10-15 as the path index.

A portion of the result generated in 204 may also be used as a trunk index to select a particular output port for forwarding the packet from multiple output ports associated with a selected trunk (step 208). The portion of the result used in 208 is generally less than the full result. The portion used in 208 may be a contiguous portion of the result generated in 204 or may be made up of non-contiguous portions of the result.

The portions of the result selected in 206 and 208 and used as a path index and as a trunk index are such that the portion used as the path index is not the same as the portion used as the trunk index. It is possible that the portion selected as the path index and the portion selected as the trunk index may have some overlapping sections. In some embodiments, the two portions are non-overlapping, for example, for a 32-bit CRC result, bits 0-15 may be used as the path index and bits 16-31 may be used as the trunk index.

For example, in embodiments supporting ECMP and trunking, a first portion of the result generated in 204 may be used in 206 as an ECMP index to select a particular ECMP path, and if the selected ECMP path is a trunk, a second portion of the result generated in 204 may be used in 208 as a trunk index to select a particular output port from the multiple ports associated with the selected ECMP path trunk. For example, for a 32-bit CRC result, bits 0-15 may be used as the ECMP index and bits 16-31 may be used as the trunk index.

Although not shown in FIG. 2, the packet may then be forwarded to the output port corresponding to the selected path and/or selected trunk port. The packet may then be transmitted from network device 100 to a next-hop device using the selected output port.

Due to the manner in which the result is generated and the manner in which portions of the generated result are used as the path index and the trunk index, the two indices are not dependent upon one another: the path index is not dependent on the trunk index and the trunk index is not dependent upon the path index. The two indices are not related to each other. This provides for enhanced diversification in the selection of paths and/or trunk ports.

The diversification in the selection of paths and/or trunk ports may be further enhanced by using multiple functions to generate the result and then using portions of the result as the path index and the trunk index. For example, instead of applying a single function, three functions may be applied: (1) a pre-function; (2) a function such as the function applied in step 204 in FIG. 2; and (3) a post-function. In this embodiment, a pre-function may be first applied to the extracted sections of the packet. The results of the pre-function are then input to the function. The results of the function are then input to a post-function. Portions of the result of the post-function are then used as the path and trunk index.

FIG. 3 depicts a simplified flowchart 300 depicting a method for generating a path index and/or a trunk index using multiple functions according to an embodiment of the present invention. In one embodiment, the processing depicted in FIG. 3 is performed by packet processor 106 depicted in FIG. 1. In other embodiments, the processing may be performed by one or more components of network device 100, including or not including packet processor 106. The processing may be performed by software executed by a processor, hardware, or combinations thereof. In the embodiment depicted in FIG. 3, three functions are applied. More or less functions may be applied in alternative embodiments.

Processing starts by extracting one or more sections of the packet to be forwarded (step 302). The processing performed in 302 may be similar to the processing performed in step 202 in FIG. 2.

A first function is then applied to sections of the packet extracted in 302 to generate a first result (step 304). Various different functions may be used as a first function such as a rotate or shift function (e.g., using a barrel shifter), a bit swap function, an invert function, an XOR function, and the like. In one embodiment, a rotate function is applied to the portion(s) of the packet selected in 302. The input to the rotate function is rotated by a certain amount to generate a rotated output value. The amount of the rotation performed may be user-configurable and may be set for the network device. Different network devices in a network environment may be configured to apply different amounts of rotations. As a result, the amount of rotation applied by one network device may be different from the amount of rotation applied by another network device. Consequently, for the same packet, the rotated result generated by one network device may be different from the rotated results generated by another network device. In one embodiment, a rotation control parameter may be set for the network device that controls that amount of rotation applied by the network device. The parameter may be stored in a memory of the network device.

A second function is then applied to the results of the first function to generate a second result (step 306). For example, if the first function was a rotate function, then the rotated result is provided as input to the second function that generates a second result. Various different functions may be used as the second function. In one embodiment, a CRC function is applied in 306 to generate a CRC result. Other functions may also be applied in alternative embodiments such as a rotate or shift function, a bit swap function, an invert function, an XOR function, and the like.

A third function is then applied to the results of the second function to generate a third result (step 308). For example, if the second function was a CRC function, then the CRC result is provided as input to the third function that generates a third result. Various different functions may be used as the third function such as rotate or shift function, a bit swap function, an invert function, an XOR function, and the like. In one embodiment, a rotate function is applied in 308. The amount of the rotation performed in 308 may be user-configurable and may be set for the network device. Different network devices in a network environment may be configured to apply different amounts of rotations. As a result, the amount of rotation applied by one network device may be different from the amount of rotation applied by another network device. Consequently, for the same packet, the rotated result generated by one network device may be different from the rotated results generated by another network device. For a network device, the amount of rotation applied by the network device in 304 may be different from the amount of rotation applied in 308.

A portion of the third result generated in 308 may then be used as a path index to select a path (e.g., ECMP path), from possibly multiple paths (e.g., multiple ECMP paths), for forwarding the packet to a next-hop device (step 310). As part of 310, an output port corresponding to the selected path may also be selected. The portion of the third result used in 310 is generally less than the full result. The portion used in 310 may be a contiguous portion of the third result generated in 308 or may be made up of non-contiguous portions of the third result.

A portion of the third result generated in 308 may be used as a trunk index to select a particular output port for forwarding the packet from the multiple output ports associated with the trunk (step 312). The portion of the third result used in 312 is generally less than the full result. The portion used in 312 may be a contiguous portion of the third result generated in 308 or may be made up of non-contiguous portions of the third result.

The portions of the result selected in 310 and 312 and used as a path index and as a trunk index are such that the portion used as the path index is not the same as the portion used as the trunk index. The portion selected as the path index and the portion selected as the trunk index may have some overlapping sections. In some embodiments, the two portions are non-overlapping, for example, for a 32-bit CRC result, bits 0-15 may be used as the path index and bits 16-31 may be used as the trunk index.

In embodiments supporting ECMP and trunking, a first portion of the third result generated in 308 may be used as an ECMP index to select a particular ECMP path, and if the selected ECMP path is a trunk, a second portion of the third result generated in 308 may be used as a trunk index to select a particular output port from the multiple ports associated with the selected ECMP path trunk. For example, for a 32-bit third result, bits 0-15 may be used as the ECMP index and bits 16-31 may be used as the trunk index.

Although not shown in FIG. 3, the packet may then be forwarded to the output port corresponding to the selected path and/or selected trunk port. The packet may then be transmitted from network device 100 to a next-hop device using the output port. Further, the scope of the present invention is not restricted to applying three functions. In alternative embodiments, the number of functions used to generate a result may be more or less functions than those depicted in FIG. 3.

It should be apparent that both the path index and the trunk index do not have to be used each time. A trunk index may need to be used only when the selected path for forwarding a packet is a trunk. Similarly, a path index may need to be used only when there are multiple paths available for forwarding the packet and a single path has to be selected. There may be no need to use a path index in network topologies where there is only one or no path from a network device for forwarding a packet. Embodiments of the present invention thus provide the flexibility of using a path index and/or a trunk index as and when appropriate.

FIG. 4 depicts a simplified block diagram of modules that may be used to implement an embodiment of the present invention. The modules in FIG. 4 may be implemented in software (e.g., program code or instructions executed by a processor), or in hardware, or combinations thereof. The software may be stored on a computer-readable medium such as memory, a disk, etc. In one embodiment, the functionality of the modules depicted in FIG. 4 may be implemented by packet processor 106. For example, the modules may be implemented as hardware components of packet processor 106. The functionality of the modules may be implemented in a programmable logic device such as a field programmable gate array (FPGA) or in an ASIC. The embodiment depicted in FIG. 4 assumes that the path index is used to select an ECMP path. The path index is accordingly referred to as the ECMP index.

In the embodiment depicted in FIG. 4, a 256-bit section 400 of a packet to be forwarded is provided to module 402 that is configured to perform a first function (e.g., a rotate function). Module 402 is configured to apply the rotate function to portion 400 and generate a 256-bit first result that is provided to a module 404 configured to perform a second function (e.g., 32-bit CRC function). The amount of rotation applied by module 402 may be preconfigured for the network device performing the processing. Module 404 is configured to apply a 32-bit CRC function to the first result and generate a 32-bit second result that is provided to module 406. Module 406 is configured to perform a third function (e.g., a rotate function) on the CRC result and generate a 32-bit third result. In case of a rotation function, the amount of rotation applied by module 406 may be preconfigured for the network device performing the processing and may be the same as or different from the amount of rotation applied by module 406. A portion of the third result (e.g., portion corresponding to bits 0-15) is used as an ECMP index to determine an ECMP path from multiple possible ECMP paths for forwarding the packet. Another portion of the third result (e.g., a portion corresponding to bits 16-31) is used as a trunk index to determine an output port for forwarding the packet from multiple output ports associated with a trunk. In one embodiment, one or more rotation control parameters may be set, for the network device, that control the amount of rotation applied by modules 402 and 406 for the network device. The parameters may be stored in a memory of the network device.

There is no limitation on the number of functions that may be used to generate a result, portions of which are then used as the ECMP index and the trunk index. For example, only one function may be used (as depicted in FIG. 2), two functions may be used (e.g., a rotate function followed by a CRC function, or a CRC function followed by a rotate function), three functions may be used (example depicted in FIGS. 3 and 4 and described above), four functions may be used, and the like.

Once the path index and the trunk index have been identified from the generated result, the indices may then be used to select a path and/or a trunk port for forwarding a packet. Various different techniques may be used to select the path and/or trunk port using a path index and a trunk index. FIG. 5 depicts a simplified flowchart 500 depicting a specific method of using a path index to select a path according to an embodiment of the present invention. The method depicted in FIG. 5 assumes that the path index is used for selecting an ECMP path and the index is thus referred to as an ECMP index. The method depicted in FIG. 5 and described below is illustrative of one embodiment of using the path index and is not meant to limit the scope of embodiments of the present invention as recited in the claims. Other methods may be used in alternative embodiments. The processing depicted in FIG. 5 may be performed by packet processor 106 depicted in FIG. 1.

As depicted in FIG. 5, a CAM lookup (e.g., using CAM 110 depicted in FIG. 1) is performed using sections extracted from the packet to be forwarded to get an index to an entry in CAM2PRAM table (e.g., CAM2PRAM 114 depicted in FIG. 1) (step 502). The sections of the packet used in 502 may or may not be the same as the sections used for generating a result as depicted in FIGS. 2 and 3. In one embodiment, a section of the header of the packet is extracted and used to perform a CAM lookup. A matching entry in the CAM yields an index pointing to an entry in the CAM2PRAM table.

An ECMP_mask value is then determined from the CAM2PRAM table entry determined from the CAM lookup performed in 502 (step 504). In one embodiment, the ECMP_mask is a 4-bit value from zero to the number of ECMP paths minus one. In order to get the number of ECMP paths and to avoid having to deal with a modulo operation (described below) using a zero base, an ECMP_Base value is then determined from the ECMP_mask determined in 504 by adding one to the ECMP_mask (step 506). In one embodiment, the ECMP_Base value is determined as follows (by adding 1 to the 4-bit number):ECMP_Base[4:0]=ECMP_mask[3:0]+4′h1//0-15→1-16 portsThe ECMP_Base identifies the total number of ECMP paths that are available for forwarding the packet from the network device.

An ECMP_Adjust value is then determined based upon the ECMP_Base determined in 506 and the ECMP or path index determined as previously described (for example, the path index determined in step 206 in FIG. 2 or in step 310 in FIG. 3) (step 508). In one embodiment, the ECMP_Adjust is determined using a modulo operation as follows:ECMP_Adjust[4:0]=ECMP Index[15:0]%ECMP_Base[4:0]

The ECMP_Adjust determined in 508 thus represents a number between 1 and the total number of available ECMP paths and used for path selection. The ECMP_Adjust is used an index to an entry in the PRAM (e.g., in PRAM 112 depicted in FIG. 1) (step 510).

The ECMP path to be used is then determined based upon the contents of the PRAM entry (step 512). In one embodiment, a forwarding identifier (FID) is determined from the PRAM entry. The particular ECMP path to be used for forwarding the packet is then determined from the FID. In one embodiment, an output port corresponding to the selected ECMP path is also determined in 512.

FIG. 6 depicts a simplified flowchart 600 depicting a method of using a trunk index to select an output port from multiple output ports associated with a trunk according to an embodiment of the present invention. The method depicted in FIG. 6 and described below is illustrative of one embodiment of using the trunk index and is not intended to limit the scope of embodiments of the present invention as recited in the claims. Other methods may be used in alternative embodiments. The processing depicted in FIG. 6 may be performed by packet processor 106 depicted in FIG. 1.

As depicted in FIG. 6, an index to an entry in the PRAM is obtained for the packet to be forwarded (step 602). In one embodiment, a section of the packet header is used to perform a lookup in the CAM. A matching entry in the CAM yields an index to an entry in the CAM2PRAM table. The CAM2PRAM table entry provides an index to an entry in the PRAM. In some embodiments, a CAM2PRAM table may not be used and a matching entry in the CAM may itself provide an index to an entry in the PRAM.

A trunk group identifier (TGID) and a PRAM forwarding identifier (PRAM FID) are then determined from the PRAM entry to which an index is obtained in 602 (step 604). In one implementation, the PRAM may be organized as a 32M×4 or 512K×64×4 memory. Each PRAM entry may include, for example, 247 bits of routing and status information, along with a 9-bit TGID (TRUNK_GROUP[8:0]), which indexes into a trunk group table. The TGID references the trunk to be used for forwarding the packet. PRAM entries sharing a trunk are programmed with the trunk's TGID.

In one embodiment, the trunk group table stores information for one or more trunks. A trunk group table may store one or more entries corresponding to trunks and the entries are addressed by TGIDs stored in the PRAM entries. The information stored for each trunk includes information identifying the current number of active ports associated with the trunk. In one implementation, each trunk group table entry stores information for a trunk and comprises a value representing the number of currently active trunk ports for the trunk. In such an implementation, when the number of active member ports in a trunk changes, the information in the corresponding trunk group table entry is updated to reflect the change. In this manner, the information in the PRAM entries for the trunk does not have to be changed. PRAM entries sharing a trunk may be programmed with the trunk's TGID, which provides an index to an entry in the trunk group table storing information for that trunk. Updates to the trunk information are made in the trunk group table, rather than in the PRAM entries.

The TGID obtained in 604 is used to access an entry in the trunk group table (step 606). The number of active ports of the trunk is then determined from the trunk group table entry accessed in 606 (step 608). In one embodiment, a 4-bit number (TRUNK_PORTS[4:0]) represents a number from zero to the number of currently active member ports of the trunk minus one. Accordingly, TRUNK_PORTS[4:0] ranges from 0 to one less than the number of currently active member ports. Thus, an adjustment may be made in 608 by adding one to the TRUNK_PORTS to obtain the number of currently active ports as follows:TRUNK_PORTS_ADJ[5:0]=TRUNK_PORTS[4:0]+5′h1TRUNK_PORTS_ADJ represents the number of active trunk ports.

An output port for forwarding the packet is then determined from the multiple ports associated with the trunk based upon the number of ports determined in 608 and the rotated trunk index (step 610). In one embodiment, this is done by first determining a Trunk_Adjust value based upon the number of active trunk ports determined in 608 and the trunk index previously obtained (as determined above in step 208 of FIG. 2 or in step 312 of FIG. 3). In one embodiment, the Trunk_Adjust value is determined as follows (assuming an 8-bit trunk index):Trunk_Adjust[5:0]=Trunk Index[15:0]%TRUNK_PORTS_ADJ[5:0]As shown above, a modulo operation is used to select one of the currently active ports of the trunk represented by Trunk_Adjust. A trunk_FID is then obtained using the PRAM FID determined in 604 and the Trunk_Adjust. In one embodiment, a 16-bit trunk_FID is obtained using a bit-wise OR operation (|) as follows:TRUNK_FID[15:0]=PRAM_FID[15:0]|Trunk_Adjust[4:0]In one embodiment, the trunk_FID points to information related to the output port from the trunk to be used for forwarding the packet. Various other techniques may also be used in alternative embodiments.

In embodiments wherein a path has to be determined from multiple paths available for forwarding the packet and the selected path is a trunk, then the processing depicted in FIG. 5 and in FIG. 6 and described above may be performed to determine a particular ECMP path and a particular output port of a trunk for forwarding a packet.

As described above, embodiments of the present invention provide techniques for generating an index (path index) for selecting a path from multiple paths available for forwarding a packet and for generating an index (trunk index) for selecting a port from multiple ports associated with a port. A result is generated and portions of the result are used as the path index and a trunk index such that the portion used as the path index is not the same as the portion used for the trunk index, although the two portions may have some overlaps. In this manner, the path index and the trunk index are not dependent on one another. Making the indices independent of each other enhances the selection diversification provided by the indices.

Embodiments of the present invention ensure that packets belonging to the same “flow” are forwarded using the same path and/or same trunk port. A “flow” may be characterized by information selected from a packet to be forwarded. For example, one or more fields selected from the header of a packet may define a flow. The definition of a flow may change from one environment to another. In one embodiment, a flow means a combination of the source and destination fields in the packet. In such an environment, two packets may be considered to belong to the same flow if they both have the same source address and destination address. For a given flow, as long as the region boundary (e.g., the number of ports configured) is fixed, the same next-hop will be chosen for packets belonging to the flow. This is useful for several networking protocol and applications. For example, for a connected TCP flow, diverting packets from the same flow to different paths/ports will introduce additional overhead due to path/port setup requirements which will degrade the performance of the network. As a result, embodiments of the present invention ensure that packets belonging to the same flow are forwarded using the same path and/or same trunk port.

Embodiments of the present invention may be used for various applications. For example, service providers and datacenters looking for ECMP and trunk diversification may use embodiments of the present invention. Embodiments of the present invention may be used in various different network topologies including single and multi-stage networks. Embodiments of the present invention provide improved diversification in the selection of paths and/or trunk ports and as a result provide improved data load balancing across paths and trunk ports in a network.

FIG. 7 depicts a multistage network 700 that may incorporate an embodiment of the present invention. FIG. 7 depicts three stages of network 700, with network device 701 belonging to a first stage, network devices 702-1, 702-2, and 702-3 belonging to a second stage, and network devices 703-1 through 703-9 belonging to a third stage. Each network device depicted in FIG. 7 may be configured to generate a result by applying a set of one or more functions to selected portion(s) of a packet, and to use portions of the result as an ECMP index and a trunk index, as previously described. Each network device may be configured to apply the same set of functions. For the embodiment depicted in FIG. 7, it is assumed that each network device applies the method depicted in FIG. 3 wherein three functions are applied to generate a result and portions of the result are used as the path index and the trunk index. The functions may include a rotate function, followed by a CRC function, followed by another rotate function, as depicted in FIG. 4.

As depicted in FIG. 7, network device 701 may receive packets belonging to multiple traffic flows, including packets belonging to flows X, Y, and Z. At network device 701, three ECMP paths (labeled 1, 2, and 3) may be available for forwarding the data flows to the next-hop. Using an ECMP index generated at network device 701, network device 701 may select ECMP path 1 for packets belonging to flow X and forward the flow X packets along ECMP path 1 to network device 702-1. Using the same technique for generating an ECMP index and a trunk index, network device 701 may select ECMP path 2 for packets belonging to flow Y and forward the flow Y packets along ECMP path 2 to network device 702-2. Using the same technique for generating an ECMP index and a trunk index, network device 701 may select ECMP path 3 for packets belonging to flow Z and forward the flow Z packets along ECMP path 2 to network device 702-3. In this manner, even though the same technique is used for generating an ECMP index, different ECMP paths are selected for forwarding packets related to flows X, Y, and Z. Further, all the packets belonging to the same flow are forwarded along the same ECMP path.

Network device 702-1 may also provide three different possible ECMP paths for forwarding packets belonging to flow X. Network device 702-1 may use the same technique for generating an ECMP index and a trunk index as used by network device 701. Based upon the ECMP index, network device 702-1 may select ECMP path 2 for forwarding packets belonging to flow X and forward the flow X packets along ECMP path 2 to network device 703-2.

Network device 702-2 may also provide three different possible ECMP paths for forwarding packets belonging to flow Y. Network device 702-2 may use the same technique for generating an ECMP index and a trunk index as used by network device 701. Based upon the ECMP index, network device 702-1 may select ECMP path 3 for forwarding packets belonging to flow Y and forward the flow Y packets along ECMP path 3 to network device 703-6.

Network device 702-3 may also provide three different possible ECMP paths for forwarding packets belonging to flow Z. Network device 702-3 may use the same technique for generating an ECMP index and a trunk index as used by network device 701. Based upon the ECMP index, network device 702-3 may select ECMP path 1 for forwarding packets belonging to flow Z and forward the flow Z packets along ECMP path 1 to network device 703-7.

As can be seen from the above example, the ECMP paths selected for forwarding a particular flow of packets at the different stages are different, even though the method used by each network device for generating the ECMP index and the trunk index is the same. For example, for packets belonging to flow X, at stage one, ECMP path 1 is selected, whereas, at stage two, ECMP path 2 is selected. This is due to the fact that the amount of rotation applied by network device 701 may be different from the amount of rotation applied by network device 702-1. As a result, the generated result after applying the three functions may be different at network device 702 from the result generated at network device 702-1. Due to this, a traffic flow may be diverted to different paths at different stages of the network while ensuring that, at a particular network device, all packets belonging to a particular traffic flow are all forwarded using the same path. This is different from conventional selection techniques, wherein the same path is likely to be selected at each network device for a traffic flow. Likewise, for packets belonging to flow Y, at stage one, ECMP path 2 is selected, whereas, at stage two, ECMP path 3 is selected. Further, for packets belonging to flow Z, at stage one, ECMP path 3 is selected, whereas, at stage two, ECMP path 1 is selected.

Accordingly, different ECMP paths are selected at different stages, thereby enhancing the diversity of the selection, using the same technique for generating path and trunk indices. Embodiments of the present invention may thus be used in a multistage network to enhance diversity in the selection of paths at the different stages. Since the same technique is used for generating the path index and the trunk index at the various network devices at the various network stages, the solution is scalable for a multistage network.

If the selected path is a trunk, then the trunk index may be used to select a particular port from the multiple ports associated with the trunk for forwarding the packet. As with selection of paths, the selected port for a particular flow of traffic may be different at different stages of the network.

Although specific embodiments of the invention have been described, various modifications, alterations, alternative constructions, and equivalents are also encompassed within the scope of the invention. For example, while embodiments have been described for using a path index to select an ECMP path from multiple ECMP paths, the path index may also be used to select other types of paths. Embodiments of the present invention are not restricted to operation within certain specific data processing environments, but are free to operate within a plurality of data processing environments. Additionally, although embodiments of the present invention have been described using a particular series of transactions and steps, it should be apparent to those skilled in the art that the scope of the present invention is not limited to the described series of transactions and steps.

Further, while embodiments of the present invention have been described using a particular combination of hardware and software, it should be recognized that other combinations of hardware and software are also within the scope of the present invention. Embodiments of the present invention may be implemented only in hardware, or only in software, or using combinations thereof.

The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. It will, however, be evident that additions, subtractions, deletions, and other modifications and changes may be made thereunto without departing from the broader spirit and scope as set forth in the claims.

Claims

1. A method comprising: applying, by a network device, one or more mathematical functions to one or more sections of a packet to generate a result value;using, by the network device, a first portion of the result value to select an equal-cost-multipath (ECMP) path from a plurality of ECMP paths for forwarding the packet from the network device; andwhen the selected ECMP path is a trunk comprising a plurality of trunk ports, using, by the network device, a second portion of the result value to select a trunk port from the plurality of trunk ports, wherein the second portion is different from the first portion, and wherein the first portion and the second portion are non-overlapping portions of the result value.

2. The method of claim 1 wherein: the one or more mathematical functions comprise a cyclic redundancy check (CRC) function; andgenerating the result value comprises using the CRC function to generate the result value.

3. The method of claim 1 wherein: the one or more mathematical functions comprise a cyclic redundancy check (CRC) function and a rotate function, wherein an amount of rotation performed by the rotate function is preconfigured for the network device; andgenerating the result value comprises using the CRC function and the rotate function to generate the result value.

4. The method of claim 1 wherein applying the one or more mathematical functions comprises: applying a first mathematical function to the one or more sections of the packet to generate a first result value; andapplying a second mathematical function to the first result value to generate the result value.

5. The method of claim 4 wherein the first mathematical function is a rotate function and the second mathematical function is a CRC function.

6. The method of claim 4 wherein the first mathematical function is a CRC function and the second mathematical function is a rotate function.

7. The method of claim 1 wherein applying the one or more mathematical functions comprises: applying a first mathematical function to the one or more sections of the packet to generate a first result value;applying a second mathematical function to the first result value to generate a second result value; andapplying a third mathematical function to the second result value to generate the result value.

8. The method of claim 7 where at least the first mathematical function or the third mathematical function is a rotate function and the second mathematical function is a CRC function.

9. The method of claim 1 further comprising: communicating the packet from the network device to a second network device using the selected ECMP path;wherein a result value generated by the second network device using the one or more sections of the packet and the one or more mathematical functions is different from the result value generated by the network device from which the packet was communicated to the second network device.

10. The method of claim 1 wherein the first portion and the second portion are each less than the result value.

11. The method of claim 1 wherein the first portion or the second portion is made of up non-contiguous sections of the result value.

12. A network device comprising: a plurality of ports; anda packet processor configurable to: apply one or more mathematical functions to one or more sections of a packet to generate a result value;use a first portion of the result value to select an equal-cost-multipath (ECMP) path from a plurality of ECMP paths for forwarding the packet from the network device; andwhen the selected ECMP path is a trunk comprising a plurality of trunk ports, use a second portion of the result value to select a trunk port from the plurality of trunk ports, wherein the second portion is different from the first portion, and wherein the first portion and the second portion are non-overlapping portions of the result value.

13. The network device of claim 12 wherein the one or more mathematical functions comprise a cyclic redundancy check (CRC) function and the packet processor is configured to use the CRC function to generate the result value.

14. The network device of claim 12 wherein the one or more mathematical functions comprise a cyclic redundancy check (CRC) function and a rotate function and the packet processor is configured to use the CRC function and the rotate function to generate the result value, wherein the amount of rotation performed by the rotate function is preconfigured for the network device.

15. The network device of claim 12 wherein applying the one or more mathematical functions comprises: applying a first mathematical function to the one or more sections of the packet to generate a first result value; andapplying a second mathematical function to the first result value to generate the result value.

16. The network device of claim 15 wherein the first mathematical function is a rotate function and the second mathematical function is a CRC function.

17. The network device of claim 15 wherein the first mathematical function is a CRC function and the second mathematical function is a rotate function.

18. The network device of claim 12 wherein applying the one or more mathematical functions comprises: applying a first mathematical function to the one or more sections of the packet to generate a first result value;applying a second mathematical function to the first result value to generate a second result value; andapplying a third mathematical function to the second result value to generate the result value.

19. The network device of claim 18 where at least the first mathematical function or the third mathematical function is a rotate function and the second mathematical function is a CRC function.

20. A network device comprising: means for applying one or more mathematical functions to one or more sections of a packet to generate a result value;means for using a first portion of the result value to select an equal-cost-multipath (ECMP) path from a plurality of ECMP paths for forwarding the packet from the network device; andmeans for, when the selected ECMP path is a trunk comprising a plurality of trunk ports, using a second portion of the result value to select a trunk port from the plurality of trunk ports, wherein the second portion is different from the first portion, and wherein the first portion and the second portion are non-overlapping portions of the result value.