Self tuning buffer allocation in a shared-memory switch
An N-port, shared-memory switch allocates a shared headroom buffer pool (Ps) for a priority group (PG). Ps is smaller than a worst case headroom buffer pool (Pw), where Pw equals the sum of worst case headrooms corresponding to each port-priority tuple (PPT) associated with the PG. Each worst case headroom comprises headroom required to buffer worst case, post-pause, traffic received on that PPT. Subject to a PPT maximum, each PPT may consume Ps as needed. Because rarely will all PPTs simultaneously experience worst case traffic, Ps may be significantly smaller than Pw, e.g., Ps<(Pw/A) where M>=2. Ps may be size-adjusted based on utilization of Ps, without halting traffic to or from the switch. If Ps utilization exceeds an upper utilization threshold, Ps may be increased, subject to a maximum threshold (Pmax). Conversely, if utilization falls below a lower utilization threshold, Ps may be decreased.
1. A switching method for a shared-memory switch comprising a plurality of ports supporting a plurality of priority levels, the method comprising:
for each port-priority tuple (PPT) associated with a priority group (PG), estimating a worst case headroom (Hw) for a connection to a peer, wherein the shared-memory switch supports a pause command to suspend traffic via the connection and wherein each worst cast headroom is indicative of a headroom buffer required to buffer PPT traffic received from the peer via the connection;
allocating a shared headroom buffer pool (Ps) shared by all PPTs in the priority group, wherein Ps is less than a worst case headroom buffer pool (Pw), and Pw is equal to the sum of each Hw corresponding to a PPT within the priority group; and
subject to a PPT maximum, permitting any particular PPT in the priority group to consume Ps as needed for traffic received after sending a pause command.
2. The method of claim 1 , wherein the PPT maximum for a particular PPT comprises the worst case headroom, Hw, for the particular PPT.
3. The method of claim 1 , further comprising:
monitoring utilization of Ps; and
adjusting Ps in accordance with the utilization.
4. The method of claim 3 , wherein adjusting Ps comprises:
responsive to detecting the utilization of Ps exceeding a high utilization threshold, increasing Ps, subject to a maximum shared headroom buffer pool (Pmax); and
responsive to detecting the utilization less than a low utilization threshold, decreasing Ps, subject to a minimum shared headroom buffer pool (Pmin);.
5. The method of claim 4 , wherein increasing Ps comprises:
determining a headroom increase corresponding to the product of Ps and an increase factor; and
subject to Pmax, increasing Ps by the headroom increase.
6. The method of claim 5 , wherein decreasing Ps comprises:
determining an un-utilized portion of Ps;
determining a shared headroom buffer pool reduction comprising a product of the un-utilized portion and a reduction factor; and
subject to Pmin, releasing a portion of Ps equal to shared headroom buffer pool reduction.
7. The method of claim 6 , wherein the low utilization threshold is 0.8, the reduction factor is 0.5, the high utilization threshold is 0.9, and the increase factor is 1.1.
8. The method of claim 4 , wherein Pmax is equal to Pw and wherein Pmin is equal to N maximum transmission units (MTUs) where N is the number of PPTs associated with the PG.
9. The method of claim 4 , wherein the shared memory switching includes at least one configuration register indicative of Ps and wherein adjusting Ps includes modifying a value of the at least one configuration register without halting packet traffic.
10. An information handling system comprising:
an embedded controller;
a plurality of ports;
a shared-memory switch;
a switching fabric, controlled by the embedded controller, interconnecting the plurality of ports; and
a computer readable medium including instructions, executable by the embedded controller, wherein the instructions, when executed by the embedded controller, cause the embedded controller to perform shared pool operations comprising:
for each port-priority tuple (PPT) associated with a priority group (PG), estimating a worst case headroom (Hw) for a connection to a peer, wherein the shared-memory switch supports a pause command to suspend traffic via the connection and wherein each worst cast headroom is indicative of a headroom buffer required to buffer post-pause traffic received from the peer via the connection;
allocating a shared headroom buffer pool (Ps) shared by all PPTs in the priority group, wherein Ps is less than a worst case headroom buffer pool (Pw), and Pw is equal to the sum of each Hw associated with a PPT of the priority group; and
permitting any particular PPT in the priority group to consume Ps as needed for post-pause traffic, subject to a PPT maximum equal to the worst case headroom Hw for the particular PPT.
11. The information handling system of claim 10 , wherein Ps is less than Pw/M, wherein M is greater than or equal to 2.
12. The information handling system of claim 10 , further comprising:
monitoring utilization of Ps; and
adjusting Ps in accordance with the utilization.
13. The information handling system of claim 12 , wherein adjusting Ps comprises:
responsive to detecting the utilization exceeding a high utilization threshold, increasing Ps, subject to a maximum shared headroom buffer pool threshold (Pmax); and
responsive to detecting the utilization less than a low utilization threshold, decreasing Ps.
14. The information handling system of claim 13 , wherein decreasing Ps comprises:
determining an un-utilized portion of Ps;
determining a pool reduction comprising a product of the un-utilized portion and a reduction factor; and
subject to Pmin, releasing a portion of Ps equal to the pool reduction.
15. The information handling system of claim 14 , wherein increasing Ps comprises:
determining a headroom increase comprising a product of Ps and an increase factor; and
subject to Pmax, increasing Ps by the headroom increase.
16. The information handling system of claim 14 , wherein releasing a portion of Ps comprises one of:
releasing the portion of Ps for burst traffic; and
increasing a shared headroom buffer pool of a different priority group; and
increasing a non-headroom buffer pool.
17. The information handling system of claim 13 , wherein the shared pool operations further comprise performing the shared pool operations of claim 13 for each of a plurality of priority groups to allocate a plurality of shared headroom buffer pools, one corresponding to each priority group.
18. The information handling system of claim 13 , wherein the priority group comprises a lossless priority group comprising a plurality of lossless levels.
19. A non-transitory computer readable medium, including instructions executable by a processor, wherein the instructions, when executed, cause the processor to perform operations, comprising:
for each port-priority tuple (PPT) associated with a priority group (PG), estimating a worst case headroom (Hw) for a connection to a peer, wherein the shared-memory switch supports a pause command to suspend traffic via the connection and wherein each worst cast headroom is indicative of a headroom buffer required to buffer post-pause traffic received from the peer via the connection;
allocating a shared headroom buffer pool (Ps) shared by all PPTs in the priority group, wherein Ps is less than a worst case headroom buffer pool (Pw), and Pw is equal to the sum of each Hw associated with a PPT of the priority group; and
permitting any particular PPT in the priority group to consume Ps as needed for post-pause traffic, subject to a PPT maximum.
20. The non-transitory computer readable medium of claim 19 , wherein the operations include:
monitoring utilization of Ps;
responsive to detecting the utilization exceeding a high utilization threshold, increasing Ps, subject to a maximum shared headroom buffer pool threshold (Pmax); and
responsive to detecting the utilization less than a low utilization threshold, decreasing Ps, subject to a minimum shared headroom buffer pool threshold (Pmin); and
wherein Ps is less than Pw/M, M is greater than or equal to 2, Hmax is equal to Hw, and Pmax is equal to Pw.