RNIC caches PTE, PTT, WQE and QP states in its SRAM. However, when the number of connections increases, the above informations are more likely to be evicted out of the RNIC cache because the cache is small, and the RNIC has to read them back from memory when processing them. Then I have a question: What is the cache replacement policy in RNIC’s SRAM cache? LRU, FIFO, or others? Thank you! I look forward to your answer.
samerka
2
Hi Xuan,
Thank you for posting the question in the Mellanox Community.
In order to provide you with such information please open a support ticket at
Thanks,
Samer
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Explanation of TX2 On-Die Cache Population Methods | 1 | 850 | August 11, 2017 | |
| Fermi Cache Architecture Cache, write policy, read policy, architecture | 6 | 3334 | August 31, 2011 | |
| Cache evict policy | 4 | 128 | July 15, 2025 | |
| Texture Cache Refilling policy | 1 | 604 | February 16, 2011 | |
| What is the write policy of L2 cache of RTX 3090 ti | 0 | 57 | August 28, 2024 | |
| About evict policy | 2 | 1492 | June 21, 2023 | |
| Is RNIC read/write data in increasing address order | 0 | 684 | June 10, 2022 | |
| Jetson TX1 Cache Policy/ System Profiler info | 2 | 952 | April 8, 2016 | |
| Adjusting L2 Persistence Cache Hit Ratio | 1 | 576 | January 18, 2024 | |
| Orin System Cache | 3 | 2368 | August 16, 2022 |