Feng Ren

Results 25 comments of Feng Ren

We have modified the code to set `attr.ah_attr.is_global=1` by default to avoid the problem of failing to establish connections in some network environments. You should also focus on the `"infiniband"`...

If you run `test_rdma` manually, all optimizations of SMART, including work request throttling, are enabled by default. You can checkout `smart_config.json`. In addition, as the test platform has only 32...

- Does you enable Hyper-Thread? Our tests assume that #thread count

P2PHANDSHAKE模式下,target 会随机分配一个端口号,显示在 I1201 14:13:27.688314 1782938 transfer_engine.cpp:146] Transfer Engine RPC using P2P handshake, listening on 100.100.100.1:16204 这一行中(即100.100.100.1:16204),这个作为 initiator 端 --segment_id 的值。同时,你不需要指定 --local_server_name 的端口号。(如果不是同机,最好不需要指定这个参数)

我现在没有很明确的思路,因为传进去的 port id 似乎是正确的,而转变到INIT阶段的传入参数也没几个: ```cpp memset(&attr, 0, sizeof(attr)); attr.qp_state = IBV_QPS_INIT; attr.port_num = context_.portNum(); attr.pkey_index = 0; attr.qp_access_flags = IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_READ | IBV_ACCESS_REMOTE_WRITE | IBV_ACCESS_REMOTE_ATOMIC; ret = ibv_modify_qp( qp,...

@1998zxn Is there any performance result about how both options affect TTFT?

We are preparing a detailed step-by-step guide of NIXL Bench. Here is a brief version: 1. Build & Install Mooncake: ``` git clone https://github.com/kvcache-ai/Mooncake sudo bash ./dependencies.sh mkdir build &&...

在图中的这个场景中,本地 mlx5_6 和远程 mlx5_3,mlx5_5,...等直接联通吗?

你可以尝试导出完整的日志记录,可能的原因包括握手阶段对端发出的错误(如 GID 选择出错等)

We expect to make it fully ready at 26Q1, and both version may coexist in about 3-6 months.