NVIDIA RoCE Congestion Control Configuration User Guide (via DMS)
Contents:
NVIDIA RoCE congestion control configuration tool (DMS) is currently supported at beta level.
TBD
All of the configuration examples below will be done using DOCA Management Service CLI, which requires DOCA DMS to be enabled.
Please refer to DOCA Management Service Guide for the instructions how to set up DOCA DMS.
<General RoCE CC overview and high-level desctiption for supported CC algorithms - DCQCN, ZTR-CC, etc.>
This section explains how to configure RoCE Congestion Control on NVIDIA BlueField and ConnectX devices.
Configuring Congestion Control Algorithms for RoCE
Verifying existing CC algorithm configuration
dmsc <flags> get --path ...
Configuration steps to enable DCQCN
dmsc <flags> set --update ...
Configuration steps to enable ZTR-CC
dmsc <flags> set --update ...
Tuning RoCE CC algorithm parameters
dmsc <flags> set --update ...
Example configuration scripts
It is possible to do multiple configuration steps at the same time by running bulk configuration update from JSON file.
The examples below show how to apply complete configuration changes to enable specific RoCE CC algorithm.
Applying end-to-end configuration for DCQCN
dmsc <flags> set --request-file req_dcqcn.json
Bulk configuration example for DCQCN:
req_dcqcn.json
{
"updates"
:
[
{
"path"
: "/interfaces/interface[name=p0]/config/mtu"
,
"value"
: 9216
,
"encoding"
: "uint"
},
{
"path"
: "/interfaces/interface[name=p0]/config/enabled"
,
"value"
: true
,
"encoding"
: "bool"
}
]
}
Applying end-to-end configuration for ZTR-CC
dmsc <flags> set --request-file req_ztrcc.json
Bulk configuration example for ZTR-CC:
req_ztrcc.json
{
"updates"
:
[
{
"path"
: "..."
,
"value"
: ...,
"encoding"
: "uint"
},
{
"path"
: "..."
,
"value"
: ...,
"encoding"
: "..."
}
]
}
This section should include recommendation to troubleshoot RoCE CC configuration.