Allen
Allen
@brminich A new clue: The buffer sent by ibv_post_send is actually not the one seen in the log, but a buffer that has been released. The request sending is serial....
``` --> client [2025-04-25 05:53:56.702] [36789] request_id:2506 alloc_buffer:0x52500190a100 buffer:"XXXXXXXX 2506 ABCDE..." [2025-04-25 05:53:56.702] [36789] request_id:2506 receive server response and release 0x52500190a100 [1745560436.702221] [node-57:36789:0] rc_verbs_ep.c:54 UCX DATA QP 0x1b7e wrid 0x9cd...
@brminich 1. Our usage method is similar to ucp_client_server, using the C++ new/delete method to dynamically allocate and release memory. This approach will cause RDMA to send old data, which...
@yosefe I tried UCX_MAX_EAGER_LANES=2 and UCX_MAX_EAGER_LANES=4, and found that the bandwidth was lower. client ``` [root@node-57 ucx-1.18.0]# UCX_MAX_EAGER_LANES=4 UCX_IB_NUM_PATHS=2 UCX_MAX_RNDV_LANES=2 UCX_NET_DEVICES=mlx5_bond_0:1 UCX_PROTO_ENABLE=y UCX_TLS=rc UCX_PROTO_INFO=y ./install-release-mt/bin/ucx_perftest 10.16.29.58 -t ucp_am_bw -s 1048576...
Hi @yosefe May I ask if this phenomenon is normal? What might it be related to?
Hello @yosefe , our RPC framework developed based on UCX is blocked in this issue. I'm not sure if the problem of iov not being able to fully utilize the...
@yosefe Thank you for your attention to my question. Strictly speaking, I use the same methods in ucp_hello_world and ucx_perftest. First, the client will use an extra OOB TCP connection...
> > Does this method not support using `ctrl-c` to interrupt the process? > > If not, should I create `ucp_listener`s that use different port for all threads? > >...
> @ivanallen 2 possible options are to extract the client address or client_id from the connection request, see https://github.com/openucx/ucx/blob/fa01cca77754edb7dd510190640c5feb9fb2b366/src/ucp/api/ucp.h#L1520C3-L1520C26 Hi @yosefe Yes, I know this API. But this can only...
> I see, so to have precise control on the client can have multiple listeners on different ports. @yosefe Thank you very much. Finally, I would like to ask why...