Processing concurrency in a network device
One or more processing operations with respect to a packet are performed at a packet processing node of a network device, the packet processing node configured to perform multiple different processing operations with respect to the packet. A first accelerator engine is triggered for performing a first additional processing operation with respect to the packet. The first additional processing operation constitutes an operation that is different from the multiple different processing operations that the packet processing node is configured to perform. The first additional processing operation is performed by the first accelerator engine. Concurrently with performing the first additional processing operation at the first accelerator engine, at least a portion of a second additional processing operation with respect to the packet is performed by the packet processing node, the second additional processing operation not dependent on a result of the first additional processing operation.
1. A method for processing packets in a network device that includes at least one packet processing node and a plurality of accelerator engines, the method comprising:
performing one or more processing operations with respect to a packet at the packet processing node, the packet processing node being configured to perform multiple different processing operations with respect to the packet;
triggering a first accelerator engine for performing a first additional processing operation with respect to the packet, wherein the first additional processing operation constitutes an operation that is different from the multiple different processing operations that the packet processing node is configured to perform;
performing the first additional processing operation by the first accelerator engine; and
concurrently with performing the first additional processing operation at the first accelerator engine, performing at least a portion of a second additional processing operation with respect to the packet by the packet processing node, wherein the second additional processing operation is not dependent on a result of the first additional processing operation.
2. The method of claim 1 , further comprising:
before the first additional processing operation is completed by the first accelerator engine, triggering a second accelerator engine for performing a third additional processing operation with respect to the packet, wherein the third additional processing operation constitutes an operation that is different from the multiple different processing operations that the packet processing node is configured to perform, and
performing the third additional processing operation on the packet by the second processing engine, including performing at least a portion of the third additional processing operation concurrently with performing at least a portion of the first additional processing operation by the first accelerator engine.
3. The method of claim 1 , further comprising, after performing the at least the second additional processing operation with respect to the packet, suspending processing of the packet at the packet processing node when no additional processing with respect to the packet can be performed at the packet processing node without a result of the first additional processing operation.
4. The method of claim 3 , further comprising:
receiving, at the packet processing node from the first accelerator engine, the result of the first additional processing operation,
in response to receiving the result of the first additional processing operation, resuming processing of the packet by the packet processing node, and
performing one or more further processing operations with respect to the packet at the packet processing node based on the result of the first additional processing operation.
5. The method of claim 1 , wherein the packet processing node includes a processing unit configured to execute computer readable instructions stored in a non-transitory memory, and wherein triggering the first accelerator engine for performing the first additional processing operation with respect to the packet includes
issuing a load command by the processing unit, wherein the load command (i) provides an indication of the first additional processing operation and (ii) requests a channel identifier (ID) corresponding to a channel used to conduct a transaction with the first accelerator engine for performing the first additional processing operation,
in response to issuing the load command, loading, at the processing unit, the channel identifier (ID) corresponding to the channel used to conduct a transaction with the first accelerator engine, and
locking the channel ID to indicate that a result of the first additional processing operation is not available for retrieval by the processing unit.
6. The method of claim 5 , wherein the load command includes, in an address field of the load command, an identifier corresponding to an entry in a profile table, wherein the entry in the profile table includes information for generating a request to the first accelerator engine, and wherein triggering the first accelerator engine further includes:
accessing the profile table based on the identifier corresponding to the entry in the profile table,
retrieving, from the entry of the profile table, information for generating the request to the first accelerator engine,
generating the request based on the retrieved information, and
sending the request to the first accelerator engine.
7. The method of claim 6 , further comprising
receiving, at the packet processing node, the result of the first processing operation performed by the first accelerator engine, and
in response to receiving the result of the first additional processing operation, releasing the channel ID to indicate that the result of the first additional processing operation is available for retrieval by the processing unit.
8. The method of claim 7 , wherein the entry in the profile table further includes an address of a memory location for storing the result of the first processing operation, and wherein the method further comprises writing the result of the first additional processing operation to the first memory location.
9. The method of claim 8 , wherein the load command to initiate triggering of the first accelerator engine is a first load command, and wherein the method further comprises
issuing, by the processing unit, a second load command to retrieve data from the first memory location, wherein the second load command includes, in an address field of the second load command, (i) the channel ID and (ii) the address of the first memory location,
determining whether the channel ID has been released, and
when it is determined that the channel ID has been released, causing data from the first memory location to be loaded by the processing unit.
10. The method of claim 9 , wherein (i) releasing the channel ID, (ii) determining that the channel ID has been released and (iii) retrieving at least a portion of the result is performed during a same clock cycle.
11. A network device, comprising
a plurality of network ports configured to receive and to transmit packets on a network;
at least one packet processing node configured to perform multiple different packet processing operations with respect to the packets, and
a plurality of accelerator engines, including at least a first accelerator engine configured to perform a first additional processing operation, wherein the first additional processing operation constitutes an operation that is different from the multiple different processing operations that the packet processing node is configured to perform,
wherein the at least one packet processing node is configured to
perform one or more processing operations on a packet;
trigger the first accelerator engine for performing the first additional processing operation with respect to the packet, and
perform at least a portion of a second additional processing operation with respect to the packet concurrently with the first processing operation performed with respect to the packet by the first accelerator engine, wherein the second additional processing operation is not dependent on a result of the first additional processing operation.
12. The network device of claim 11 , wherein the packet processing node is further configured to:
before the first additional processing operation is completed by the first accelerator engine, trigger a second accelerator engine, of the plurality of accelerator engines, for performing a third additional processing operation on the packet, wherein the third additional processing operation constitutes an operation that is different from the multiple different processing operations that the packet processing node is configured to perform, and
wherein at least a portion of the third additional processing operation is performed by the second accelerator engine concurrently with at least a portion of the first additional processing operation performed by the first accelerator engine.
13. The network device of claim 11 , wherein the packet processing node is further configured to, after performing the at least the second additional processing operation with respect to the packet, suspend processing of the packet when no additional processing on the packet can be performed at the packet processing node without a result of the first additional processing operation.
14. The network device of claim 13 , wherein the packet processing node is further configured to
receive the result of the first additional processing operation from the first accelerator engine,
resume processing of the packet, and
perform one or more further processing operations with respect to the packet based on the result of the first additional processing operation.
15. The network device of claim 11 , wherein the packet processing node includes a processing unit configured to execute computer readable instructions stored in a non-transitory memory, and wherein the packet processing node is configured to
initiate triggering of the first accelerator engine by issuing a load command from the processing unit, wherein the load command (i) provides an indication of the first additional processing operation and (ii) requests a channel identifier (ID) corresponding to a channel used to conduct a transaction with the first accelerator engine for performing the first additional processing operation, and
in response to issuing the load command
load, at the processing unit, the channel identifier (ID) corresponding to the channel used for conducting a transaction with the first accelerator engine for performing the first additional processing operation, and
lock the channel ID to indicate that the result of the first additional processing operation is not available for retrieval by the processing unit.
16. The network device of claim 15 , wherein the load command issued by the processing unit includes, in an address field of the load command, an identifier corresponding to an entry in a profile table, wherein the entry in the profile table includes information for generating a request to the first accelerator engine, and wherein triggering the first accelerator engine for performing the first additional processing operation on the packet includes:
accessing the profile table based on the identifier corresponding to the entry in the profile table,
retrieving, from the entry of the profile table, information for generating the request to the first accelerator engine,
generating the request based on the retrieved information, and
sending the request to the first accelerator engine.
17. The network device of claim 16 , wherein the packet processing node is further configured to
receive the result of the first processing operation performed by the first accelerator engine, and
in response to receiving the result of the first processing operation, release the channel ID to indicate that the result of the first additional processing operation is available for retrieval by the processing unit.
18. The network device of claim 16 , wherein the entry in the profile table further includes an address of a memory location for storing the result of the first processing operation, and wherein the packet processing node is further configured to write the result of the first processing operation to the first memory location.
19. The network device of claim 15 , wherein the load command to initiate triggering of the first accelerator engine is a first load command, and wherein
the processing unit is further configured to issue, a second load command to retrieve data from the first memory location, wherein the second load command includes, in an address field of the second load command, (i) the channel ID and (ii) the address of the first memory location, and wherein
the packet processing node is further configured to
determine whether the channel ID has been released, and
when it is determined that the channel ID has been released, cause data from the first memory location to be loaded to the processing unit.
20. The network device of claim 19 , wherein the packet processing node is configured to (i) release the channel ID, (ii) determine that the channel ID has been released and (iii) cause data from the first memory location to be loaded to the processing unit during a same clock cycle.