IBUtils2 Utility Release Notes v2.8.0
NVIDIA UFM Cyber-AI Documentation v2.6.1-4

Changes and New Features History

v2.7.0

Link Speed

Added support for NDR InfiniBand link speed.

Fat Tree Topology

Enabled a new Fat Tree Topology validation tool.

Virtualization Stage

Redesigned the Virtualization stage to asynchronous mode.

AGUID

AGUID stage is disabled by default.

To enable it, use the '--aguid' parameter.

db_csv

db_csv will now contain information about connected ports only.

v2.6.1

ibdiagnet

  • Enabled Adaptive Routing validation

  • Performance improvement of routing checking

  • Added new counters and diagnostic information:

    • SHARP: “SAT” (“Streaming Aggregation”) counters

    • PHY: Maximum PLR per second field in PPCNT

    • PHY: SLLM register

    • PHY: New SymbolBER thresholds

    • Adaptive Routing: PortARTrails counter

  • Added support for SHARP security (AMKEY)

  • Enabled reporting port counter differences when using "–pm_pause_time" (PM_DELTA section in db_csv)

  • Added Dragonfly+ Topology Validation (--dfp, --dfp_opt [<max_cas>] )

  • Added the option to report the version on the screen and in log file

ibtopodif

Added support for “stable” names of RDMA devices based on PCI/slot/function location.

ibnetdiscover dump file (created by ibdiagnet)

Now the dump file includes Virtual Port info

v2.5.1

Rail Optimized Topology Validation

Checks links between compute nodes and leaf switches to provide rail optimized topology (--rail_validation, --rail_validation_opt [<regex>] ).

Service Level

Customization SL for GMP MADs in ibdiagnet (--sl).

General

  • Added output similar to the results of the "ibnetdiscovery"

  • ibdiagnet: Disabled the output by default for AR/FDBS files in routing stage

  • Moved SHARP performance counters to db_csv

  • Dumped by default FEC_MODE section in db_csv

  • Improved AR connectivity check

v2.4.0

General

  • Flexible output control options (--enable_output , --disable_output, --path)

  • Discovery only mode (–discovery_only)

  • Support for MLNX Congestion Control counters

  • “Fabric Summary” is enabled by default.

  • Dates and versions in “nodes_info” file are printed in human-readable format.

  • Dump files include ibdiagnet version and command line parameters.

  • Added split mode to IBNL for “InfiniBand Smart Director Switches” (CS8500)

Performance Improvements

The performance of following steps in ibdiagnet have been improved:

  • Routing validation

  • Network discovery

  • Virtualization stage

  • Dump creation

Version

Tool

Parameter Name

Status

Description

2.7.0

ibdiagnet

--ft

New

Enables Fat Tree Topology Validation (default - disabled)

--aguid

New

Collects AGUIDs

--enable_spst

Removed

SPST mode is enabled by default. Option was deprecated in 2.6.1

2.6.1

ibdiagnet

--smp_window

Upper limit and default are changed

Max: 256; default: 16

--gmp_window

Upper limit and default are changed

Max: 16384; default: 256

--am_key

New

Specifies default AMKEY for the fabric

--am_key_file

New

Specifies the path to file (AMKEY per GUID)

--smdb

New

Specifies the path to OpenSM SMDB file (required for Adaptive Routing & Dragonfly+ Topology validation)

--ber_threshold_table

New

Specifies the path to BER thresholds table file (BER per FEC)

--create_ber_threshold_table

New

Creates template file of BER threshold table

--enable_spst

Deprecated

SPST mode is enabled by default

--dfp

New

Enabling DFP Topology Validation (default - disabled)

--dfp_opt

New

Specifies comma separated DFP Topology Validation options

dfp_opt

<max_cas>

-

Specifies max number of CAs for “Root” switch in Dragonfly+ island (default: 1).

This parameter is mutual exclusive with --smdb

ibdmchk

--FAR

New

Adds support for input FAR file

2.5.1

ibdiagnet

--sl

New

Specifies the SL to be used (default=0)

--rail_validation

New

Enabling Rail Optimized Topology Validation (default - disabled)

--rail_validation_opt

New

Specifies comma separated Rail Optimized Topology Validation options

--clear_congestion_counters

New

Dumping Congestion Counters and clearing them

--fec_mode

Deprecated

FEC_MODE section will be dump to "db_csv" by default

--rail_validation_opt

<regex>

-

Specifies regular expression to filter HCA nodes from reports. To be applied to HCAs node descriptions

2.4.0

ibdiagnet

--enable_output

New

Enable creation of specific dump file

--disable_output

New

Disable specific dump file

--path

New

Set custom path for specific dump file

--discovery_only

New

Discover IB fabric, save topology information into “db_csv” file and exit

--smp_window

Upper limit is changed

New max is 128

--gmp_window

Upper limit is changed

New max is 8192

ibtopodiff

--ibnl_dir

New

Set path for IBNL files

© Copyright 2023, NVIDIA. Last updated on May 23, 2023.