bytedance / bytedance/InfiniStore
rdma_read/write performance review
Open
- Dominant language
- C++
- Stars
- 437
- Forks
- 44
- PR merge metrics
- No merged PRs in 30d
Description
```python benchmark.py --block-size 32 --size 128 --rdma```
when batch is big (> 4096), it takes around 1x ms. (gcc -O0)
this is too slow:
rdma_write:
mm->allocate add batch allocate, if it could allocate 4090 blocks one time. it may improve performance.
rdma_read:
TODO:
Contributor guide
Assessment
This issue has not been assessed yet.