Skip to content

Fix: umbp qp per transfer default value & add logs #384

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
TianDi101 merged 3 commits into from
Jun 11, 2026
Merged

Fix: umbp qp per transfer default value & add logs #384

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
95ae019
add log
TianDi101 Jun 10, 2026
b91b0fd
change default qp per transfer to 4
TianDi101 Jun 10, 2026
4f2398c
delete extra comment
TianDi101 Jun 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 25 additions & 2 deletions src/application/transport/rdma/providers/ibverbs/ibverbs.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,10 @@
// SOFTWARE.
#include "mori/application/transport/rdma/providers/ibverbs/ibverbs.hpp"

#include <cerrno>
#include <cstring>
#include <stdexcept>

#include "mori/application/utils/check.hpp"
#include "mori/utils/mori_log.hpp"
namespace mori {
Expand Down Expand Up @@ -69,7 +73,13 @@ RdmaEndpoint IBVerbsDeviceContext::CreateRdmaEndpoint(const RdmaEndpointConfig&
endpoint.ibvHandle.compCh = config.withCompChannel ? ibv_create_comp_channel(context) : nullptr;
endpoint.ibvHandle.cq =
ibv_create_cq(context, config.maxCqeNum, NULL, endpoint.ibvHandle.compCh, 0);
assert(endpoint.ibvHandle.cq);
if (!endpoint.ibvHandle.cq) {
MORI_APP_ERROR(
"ibv_create_cq failed: errno={} ({}); dev={} max_cqe={} dev_max_cqe={} cqs_in_pool={}",
errno, strerror(errno), GetRdmaDevice()->Name(), config.maxCqeNum,
deviceAttr->orig_attr.max_cqe, cqPool.size());
throw std::runtime_error("ibv_create_cq failed: " + std::string(strerror(errno)));
}
Comment on lines +76 to +82

// TODO: should also manage the lifecycle of completion channel && srq
if (config.withCompChannel)
Expand All @@ -92,7 +102,16 @@ RdmaEndpoint IBVerbsDeviceContext::CreateRdmaEndpoint(const RdmaEndpointConfig&
},
.qp_type = IBV_QPT_RC};
endpoint.ibvHandle.qp = ibv_create_qp(pd, &qpAttr);
assert(endpoint.ibvHandle.qp);
if (!endpoint.ibvHandle.qp) {
MORI_APP_ERROR(
"ibv_create_qp failed: errno={} ({}); dev={} port={} max_send_wr={} max_recv_wr={} "
"max_send_sge={} max_cqe={} dev_caps(max_qp_wr={} max_qp={} max_cqe={}) qps_in_pool={}",
errno, strerror(errno), GetRdmaDevice()->Name(), config.portId, qpAttr.cap.max_send_wr,
qpAttr.cap.max_recv_wr, qpAttr.cap.max_send_sge, config.maxCqeNum,
deviceAttr->orig_attr.max_qp_wr, deviceAttr->orig_attr.max_qp,
deviceAttr->orig_attr.max_cqe, qpPool.size());
throw std::runtime_error("ibv_create_qp failed: " + std::string(strerror(errno)));
}
Comment on lines +105 to +114
endpoint.handle.qpn = endpoint.ibvHandle.qp->qp_num;

if (config.enableSrq)
Expand Down Expand Up @@ -208,6 +227,10 @@ IBVerbsDevice::~IBVerbsDevice() {}

RdmaDeviceContext* IBVerbsDevice::CreateRdmaDeviceContext() {
ibv_pd* pd = ibv_alloc_pd(defaultContext);
if (!pd) {
MORI_APP_ERROR("ibv_alloc_pd failed: errno={} ({}); dev={}", errno, strerror(errno), Name());
throw std::runtime_error("ibv_alloc_pd failed: " + std::string(strerror(errno)));
}
Comment on lines 229 to +233
return new IBVerbsDeviceContext(this, pd);
}

Expand Down
3 changes: 2 additions & 1 deletion src/umbp/distributed/pool_client.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,7 @@
#include <unordered_map>

#include "mori/io/backend.hpp"
#include "mori/utils/env_utils.hpp"
#include "mori/utils/mori_log.hpp"
#include "umbp/common/env_time.h"
#include "umbp/distributed/master/master_metrics.h"
Expand Down Expand Up @@ -277,7 +278,7 @@ bool PoolClient::Init() {
io_engine_ = std::make_unique<mori::io::IOEngine>(config_.master_config.node_id, io_cfg);
mori::io::RdmaBackendConfig rdma_cfg;

rdma_cfg.qpPerTransfer = 16;
rdma_cfg.qpPerTransfer = mori::env::GetPositiveIntOr("MORI_UMBP_QP_PER_TRANSFER", 4);
rdma_cfg.enableTransferChunking = true;
rdma_cfg.numNicsPerTransfer = 4;
io_engine_->CreateBackend(mori::io::BackendType::RDMA, rdma_cfg);
Expand Down
Loading