Professional Documents
Culture Documents
MLNX-OS - IB - Version - 3.9.0900 - Release Notes PDF
MLNX-OS - IB - Version - 3.9.0900 - Release Notes PDF
MLNX-OS - IB - Version - 3.9.0900 - Release Notes PDF
Mellanox Technologies
350 Oakmead Parkway Suite 100
Sunnyvale, CA 94085
U.S.A.
www.mellanox.com
Tel: (408) 970-3400
Fax: (408) 970-3403
Mellanox®, Mellanox logo, 25 Is The New 10®, CoolBox®, InfiniScale®, LinkX®, Mellanox Care®, Mellanox CloudX®, Mellanox
NEO®, Mellanox NVMEDirect®, Mellanox OpenCloud®, Mellanox Open Ethernet®, Mellanox Spectrum®, Mellanox TuneX®,
Mellanox Virtual Modular Switch®, MetroDX®, MetroX®, MLNX-OS®, ONE SWITCH. A WORLD OF OPTIONS®, Open Ethernet
logo, Spectrum logo®, StoreX®, Switch-IB®, SwitchX®, TestX®, TuneX®, UFM®, Unbreakable-Link®, Virtual Protocol
Interconnect®, Voltaire®, and Voltaire logo are registered trademarks of Mellanox Technologies, Ltd.
For the complete and most updated list of Mellanox trademarks, visit http://www.mellanox.com/page/trademarks.
Date Description
June 4, 2020 First release of this software version.
1 Introduction
This document is the Mellanox MLNX-OS® Release Notes for InfiniBand.
MLNX-OS is a comprehensive management software solution that provides optimal perfor-
mance for cluster computing, enterprise data centers, and cloud computing over Mellanox
Switch-IB®, and Spectrum® IC families. The fabric management capabilities ensure the highest
fabric performance while the chassis management ensures the longest switch up time.
The MLNX-OS documentation package includes the following documents:
• User Manual provides general information about the scope, organization, and command
line interface of MLNX-OS as well as basic configuration examples.
• Release Notes provide information about the supported platforms, changes and new
features, and reports on software known issues as well as bug fixes.
The Mellanox Community also offers useful end-to-end and special How To guides at:
http://community.mellanox.com/community/solutions.
QM8700 Mellanox Quantum 40-port QSFP56 HDR (200Gb/s) InfiniBand smart switch x86
TQ8200 Quantum MetroX®-2 40km long haul 100Gb/s, 2 long haul QSFP28 ports, 8 standard x86
HDR ports
TQ8100 Quantum MetroX®-2 10km long haul 100Gb/s, 2 long haul QSFP28 ports, 8 standard x86
HDR ports
SB7800 Switch-IB® 2 36-port EDR (100Gb/s) 1U InfiniBand dual core switch x86
SB7700 Switch-IB 36-port EDR (100Gb/s) 1U InfiniBand dual core switch x86
SB7780 Switch-IB 36-port EDR (100Gb/s) 1U InfiniBand dual core IB router x86
CS7500 Switch-IB Family 648-port EDR (100Gb/s) InfiniBand switch x86
CS7510 Switch-IB Family 324-port EDR (100Gb/s) InfiniBand switch chassis x86
CS7520 Switch-IB Family 216-port EDR (100Gb/s) InfiniBand switch chassis x86
The minimum required firmware versions for ConnectX-6 and Mellanox Quantum are
20.25.1532 and 27.2000.1260 respectively.
When using Mellanox AOC cables that are longer than 50m, one VL must be used to
achieve full wire speed.
The following table presents the connectivity matrix, between Mellanox Quantum based
switches, ConnectX-6 HCA, and the cables.
Table 4 - Switch to Switch Connectivity
Switch Switch Cable
H cable DAC H cable AOC HDR DAC HDR AOC EDR DAC EDR AOC
Y cable DAC Y cable AOC HDR DAC HDR AOC EDR DAC EDR AOC
1. This is currently not supported. For further information see the HDR limitation in table Table 6, “Mellanox Quan-
tum Supported Link Speed”.
Release 3.9.09xx
MAC Added support for MAC masking in log messages.For more information, see "Logging"
section in the user manual.
UFM Added support for IPv6 in UFM Agent.
General Bug fixes (see Section 5, “Bug Fixes,” on page 19).
Release 3.9.0606
TQ8100 Added ES-level support for TQ8100 system.
TQ8200 Added ES-level support for TQ8200 system.
General Bug fixes (see Section 5, “Bug Fixes,” on page 19)
Release 3.9.0450
Configuration Added support for automated configuration file backup.
Management
Release 3.9.0300
Cables Added support for the following OPNs:
• MMA1L30-CM MMA1T00-HS MCA1J00-H003E/4E MCA7J50-H00XX
DCQCN Added support for DCQCN Congestion Control.
Link Added support for link-negotiated credit size.
Mellanox Scalable SHARP (SAT) is at GA level.
Hierarchical
Aggregation and *SAT: Streaming Aggregation Tree
Reduction Protocol
(SHARP)™
Security SSH Login Notification now displays the following information after authentication:
• Last successful and unsuccessful login date/time
• Number of unsuccessful logins since last successful login
• Changes to user's account since last login (password, capability)
• Location of last successful and unsuccessful login (terminal or IP)
• Number of total successful logins since last X days
Security Upgraded OpenSSH version to 8.0p1.
Split Added ability to MCS8500 systems to configure split-ready profile and split ports using
CLI.
General Bug fixes (see Section 5, “Bug Fixes,” on page 19)
Release 3.8.2204
Link Speed Added QDR/FDR support in Mellanox Quantum switch systems when using optical
cables of up to 30m.
Note: QDR speed is only supported when using FDR cables. For a list of supported
cables, see “InfiniBand Known Issues” on page 15.
Cables HDR Active Copper Cables are now supported between two switches.
General Bug fixes (see Section 5, “Bug Fixes,” on page 19)
Release 3.8.2102
NTP Server Added new CLI commands that allow the user the block the switch's ability to function
as an NTP server. For more information, see NTP commands in "NTP and Clock"
section in the user manual.
Syslog UDP/TCP The"crypto certificate system-self-signed regenerate" command was added with the
option to specify the certificate CA basic constraints flag. For more information see "
Cryptographic (X.509, IPSec) and Encryption" section in the user manual.
Subnet Manager Added additional options for IB partition's member configuration (all-cas, all-routers,
all-switches, all-vcas) to better align with keywords that are supported by opensm when
configuring IB partitions. For more information see "SM Commands" section in the
user manual.
SHARP Added GA-level support for Mellanox Scalable Hierarchical Aggregation and
Reduction Protocol (SHARP)™ v1 technology.
SHARP Added Beta-level support for Mellanox Scalable Hierarchical Aggregation and
Reduction Protocol (SHARP)™ v2 technology.
Release 3.8.2004
BER Added port Bit Error Rate (BER) monitoring for Mellanox Quantum-based switch
systems.
Cables Removed PLR from active cables longer than 30m.
CS8500 Added thermal algorithm support to CS8500.
InfiniBand Interfaces Added FDR support in Mellanox Quantum switch systems when using optical cables of
up to 30m.
Link Down Added support for Link Down Reasoning.
Reasoning
Logging Added support to send syslog messages based only on a given regex.
For more information, see “Logging Commands” section in the user manual.
Mellanox Scalable [Beta-level] Mellanox Scalable Hierarchical Aggregation and Reduction Protocol
Hierarchical (SHARP)™ technology improves the performance of MPI operations by offloading
Aggregation and collective operations from the CPU to the switch network, and by eliminating the need
Reduction Protocol to send data multiple times between endpoints.
(SHARP)™
Running-config Added version information to the show running-config.
Subnet Manager Added support to enable calculation of missing routes. See “ib sm calculate-missing-
routes” command in the MLNX-OS User Manual.
General Added support to prevent disk from running out of space due to sysdump files.
PLR Added support to the Physical Layer Retransmission (PLR) functionality for HDR
speed.
Link Up Speed Link up time improvements. The link up time is up to 60 seconds.
General Bug fixes (see Section 5, “Bug Fixes,” on page 19)
Release 3.8.1206
Switch Systems Added GA level support for HDR (200Gb/s) capable 800-port InfiniBand switch
system (CS8500).
Chassis HA Added IPv6 support for Chassis HA. For further information, see the “chassis ha bipv6”
and “show chassis ha” commands in the user manual.
Release 3.8.1102
Switch Systems Added beta-level support for HDR (200Gb/s) capable 800-port InfiniBand switch
system (CS8500).
Switch Systems Added GA-level support for QM8700 switch system.
InfiniBand Interfaces Added GA level support for HDR rate speed for Mellanox Quantum-based switch
systems.
Release 3.8.1000
JSON Added support for running a single HTTP request for both authentication and JSON
request.
IPv6 Added IPv6 support for RADIUS and Syslog.
WebUI Removed a few insecure parameters of the HTTP header.
Cables Added support for HDR AOC H-cables MFS1S90-H0xxE
Cables Added support for HDR copper cables MCP1650-H00xxxx
Cables Added support for HDR Y splitter copper cable MCP7H50-H001R30.
Docker Added the option to mount USB to a docker container.
Docker Enhanced dockers performance with Overlay2 driver.
Telemetry Added support for deleting telemetry statistics files.
CLI Added Ctrl-w key shortcut support to the CLI.
See section “CLI Shortcuts” in the User Manual.
Breakout Cables Added ability for QTM8700 system to configure split-ready mode and break-out cable
configuration using MADs.
Breakout Cables Confirmation for split-port has changed to an all-caps “YES”.
Release 3.8.0994
System Management Added alpha level support for HDR rate speed on both internal and external links.
Release 3.7.1134
General Bug Fixes
General Bug Fixes
Release 3.7.1086
System Management Added alpha level support for 40-Port HDR / 80-Port HDR100 managed Switch
System (MCS8500). This release provides the following functionalities:
• Two management slave/master modules
• Boot over Out-of-Band (OOB) ports
• EDR rate speed on both internal and external links
• The management chassis includes leaf and spine hot swapping
Release 3.7.1060
Link speed Added alpha-level support for HDR speed when using optical cables only.
Release 3.7.1000
General Upgraded to Linux kernel 4.15/6.
JSON Added JSON API support for all legacy commands.
Security Added support for remote server only in Authentication, Authorization and Accounting
(AAA) fallback.
See section “Authentication, Authorization and Accounting” in the user manual.
Security Added support for CSR certificate.
Security Added support for additional SNMP encryption cipher.
See section “SNMP” in the User Manual.
System Profile Added ability for QTM8700 Mellanox Quantum-based systems to reboot in a split-
ready mode optimized to configure twice the number of ports exposed to IB utilities.
See the parameter “split-ready” in the command “system profile”.
System Profile Added ability to set number of adaptive routing groups in system profile.
See “system profile” command in user manual.
SM Added support for OpenSM 5.2.0 which provides HDR support.
Telemetry Added support for Top Talker.
See section “Telemetry” in the User Manual.
Systems Added EDR-speed GA-level support for the QM8700 switch system.
Release 3.6.8008
JSON Added support for additional JSON commands.
See Appendix “Show Commands Not Supported by JSON.”
Linux Dockers Added ability to start a container immediately as well as after system initialization with
one CLI command.
See parameters “now-and-init” and “now-and-data-path-ready” in the command “start”
under the “Linux Docker” section in the User Manual.
Linux Dockers Added option to limit docker containers CPU and memory usage.
See command “start” in the User Manual.
Linux Dockers Added ability to monitor container statistics.
See command “show docker stats” in the User Manual.
Linux Dockers Added support for sending an SNMP health trap upon container failure.
Security Enhanced default configuration for switches to be more secure so that, by default,
HTTP is disabled, HTTPS is enabled, the lowest SSH version allowed is 2, and SSH
security strict mode is enabled.
SNMP Added a new SNMP notification to the MELLANOX-POWER-CYCLE MIB called
mellanoxPowerCyclePlannedReload (OID 1.3.6.1.4.1.33049.10.1.1.3.1) to support
planned reload.
Subnet Manager Added support for scatter port.
See section “Scatter Ports” in the User Manual.
Subnet Manager Added support for setting GUID order in routing table.
See section “GUID Routing Order” in the User Manual.
Subnet Manager Added support for bulk application of SM configurations.
See section “Bulk Update Mode” in the User Manual.
WebUI Added GA level support for Linux Dockers.
See Setup > Docker in the WebUI.
Release 3.9.09xx
MAC From this release on, the first two bytes of the MAC address in the log will be marked
with asterisks for security purposes. This may be disabled using the no form of the
“logging mac masking” command. See “Logging” section in the user manual for more
information.
Release 3.8.2004
Security Admin and Monitor passwords must be typed upon initial configuration. Default
Enhancements passwords will no longer be created automatically.
As of the September release of version 3.8.2000, California law SB-327 will be
enforced. The new law creates some limitations with respect to the use of the default
password.
Currently, the user is not required to set the default user and password for “Admin” and
“Monitor.” The user can skip the credentials wizard when setting up a new switch and
have the following default User and Password:
Admin: User & Password = admin
Monitor: User & Password = monitor
To comply with the California law SB-327 regulations, the user will now have to enter
the password manually, as part of the initial wizard and will not be allowed to skip this
step. Nevertheless, the user will be allowed to manually write in the default user name
and password (admin/admin or monitor/monitor)
The changes are implemented in a manner that minimizes the impact on the automation
processes so that Zero Touch Provisioning (ZTP) will continue to work as usual and will
not be affected by the new regulation.
XPL API As of the September release of software version 3.8.2000, the XML user accounts will no
Deprecation longer be supported and the XML gateway will be closed. Access through XML will no
longer be available.
Interfaces will only be available through SNMP and JSON.
Backward When performing downgrade or role-back from version 3.8.2000 to an older version, the
Compatibility MGMT slave will require additional reboot to complete initialization.
Web UI Re-designed interface to enhance user experience.
4 Known Issues
The following sections describe Mellanox MLNX-OS® known issues in this software release
and possible workarounds.
For hardware issues, please refer to the switch support product page.
9. Linux Dockers When running “configuration text apply” (with Run the init configuration
“docker no shutdown”), a container that is from the CLI session.
configured as init, may run immediately
(instead of waiting for next boot).
10. Linux Dockers If two docker images are installed, both from Delete older latest image
same distribution and both chosen as “latest”, or download while
the command “show docker images” may specifying a version.
display image name and image version as
“none”.
11. Logging The warning “[pm.WARNING]: snapshots and N/A
sysdumps are on separate partitions; space
constraints not thoroughly enforced on
sysdumps” may appear if operating with an
encrypted file system. This warning may be
safely ignored.
12. Logging When switch clock has an earlier time N/A
configured from the certificate creation, the
following error may appear in the log and can
be safely ignored: “[mgmtd.ERR]:
md_cert_validate_new_cert_value(),
md_cert.c:3388, build 1: Return status 512
from openssl verify!”
13. Management Switch systems may have an expired HTTPS Generate a new certificate
Interfaces certification. by changing the hostname.
14. Management Consecutive hostname modification is not Wait 25 seconds before
Interfaces supported. reattempting to modify the
hostname.
15. Management Speed of mgmt0 interface is shown as N/A
Interfaces “UNKNOWN” when working with VM.
16. SNMP Request timeout should be set to at least 20 N/A
seconds since initial table calculation requires
time.
17. Virtual Machine Virtualization connection might fail when Connect through text.
trying to connect through graphics.
18. Virtual Machine For volume fetch, using a USB drive formatted Use EXT3 USB format.
with VFAT causes errors in the log and may
require additional reboot for the USB to be
registered for virtual machine volume usage.
19. WebUI Interactive CLI commands cannot be executed N/A
via WebUI.
20. WebUI Importing a configuration text file with Import the configuration
commands that only get enabled after running text file through the CLI.
other commands is not possible through the
WebUI.
21. WebUI Reversing the time clock can result in WebUI Clear the graphs data after
graphs’ corrupted data. setting the clock.
22. WebUI Enabling/disabling HTTPS while connected Refresh the page or
via HTTP to the WebUI may result in navigate back using the
temporary loss of connection to the webpage. browser’s back button.
17. Mellanox Quantum Running mstdump on Mellanox Quantum Reload the switch.
ASIC from a host (on MFT versions 4.12.0-
15 or older) followed by split-port
operations, may cause the ASIC to not
operate properly.
18. Mellanox Quantum The following features are not currently N/A
supported on Mellanox Quantum-based
systems:
• Congestion Control
• IB Router
• Signal Degradation Monitoring
• Telemetry Protocol
19. SHARP On rare occasions and under high SHARP Disable trimming in
load, switch SHARP operation might get Aggregation Manager
stuck. settings.
20. SHARP (SAT) Data VLs buffer may be lack of credits due Use SHARP (SAT)
to an overload of small size SHARP (SAT) message size of at least
messages (<32KB). 32KB.
5. IB Router IB router packets arrive with wrong VL when Use either SL2VL or
crossing VL2VL and SL2VL. VL2VL.
5 Bug Fixes
The following table lists the latest Mellanox MLNX-OS® bug fixes.
Revision 3.9.09xx
1. Logging Default configuration of log rotation criteria "size" appears in "show running-
config".
2. Switch Management In version 3.9.06xx, when more than 1000 DNS resolutions occur via DHCP,
the switch may get stuck.
3. TQ8xxx Systems, TQ8100 and TQ8200 systems show a second management interface. This
Management second interface can be ignored.
Interfaces
Revision 3.9.0606
4. General Management When "SSH server login record-period" is set to 30 days and the successful
login count is higher than 20K, a delay may be experienced in initializing
SSH, the console, or web sessions.
5. Link On rare occasion when link is flapping or toggle by the user, the switch may
hang.
6. Logging Logging to remote host in WELF format is not working.
7. SHARP (SAT) Traffic loss may be experienced during a spine failover, when two SHARP
(SAT) flows are enabled.
8. SNMP In case of power supply cable removal, the power supply is still accessible to
SNMP requests.
9. SNMP, MIB On systems with fixed PSUs (where PSUs cannot be removed), when one PSU
is down, SNMP OID 1.3.6.1.2.1.99 fails.
Revision 3.9.0450
1. Link EDR link does not come up when using MMA1L30-CM modules with
SB7800 switches.
2. Management Inter- The IPv6 address of management interfaces disappears when link flips, if it is
faces, IPv6 a static IP.
3. SHARP (SAT), 2 Running 2 flows in parallel is currently not functional in SHARP (SAT).
Flows
4. SM HA Some online nodes are mistaken by the SM HA as being offline, when the
hostnames are only differentiated by a '.' instead a '-'.
Revision 3.9.0300
5. Auto-negotiation, As the switch does not send auto-negotiation indication, after resetting/power
HCA cycling a ConnectX-6 HCA, some HCAs get stuck in "polling" state.
6. Link Up The switch gets stuck while conducting force link up with too many PLR
retries.
24. Logging In case multiple identical log messages are sent in a short period of time, the
aggregation may not work properly.
25. Logging Upon system deinitialization, the following error may appear in log:
"[unk.ERR]: unk: SDK LOG MESSAGE: [FDB_FLOOD_DB"
This error can be safely ignored.
Revision 3.8.2032
1. Adaptive Routing Packets transmitted to the wrong output port due to misconfiguration of the
packet classification decision in the switch forwarding database cache key that
caused both AR eligible packets and AR ineligible packets to hit the same
cache entry.
2. Cables High BER when using optical module with module firmware older than
37.50.316.
Revision 3.8.2004
3. Cables The bandwidth on MFS1S00-H050E cables is 99Gb/s and on MFS1S00-
H100E cables is 67Gb/s when connecting at HDR speed to a HDR switch.
4. Chassis Management Fan speed changes too often on the system.
5. Copper Cables Link issues might occur if 1.5m copper cables and longer are not connected to
the middle ports of the Mellanox Quantum switch.
6. CLI The CLI command "show interfaces ethernet [<inf>] transceiver brief" can fail
to execute if a cable has the "#" character in its cable information data.
7. CS8500 On CS8500 switch systems, when entering (or powering on) a second manage-
ment module to a director system, a few errors may appear in the log for sev-
eral seconds. These errors can safely be ignored.
8. Error Flow On CS7500 switch systems, after takeover, spine1 may remain in "power on"
state instead of being in "ready" state.
9. JSON When JSON API was overloaded with requests, the request queue would
sometimes get to its maximum capacity.
10. SNMP SNMP through inband stops responding when the management interface is
down.
11. Software Manage- Switch may self-reboot 49 days after upgrading to version 3.8.1206 or higher.
ment
12. Web UI Switch administrator can pull files through the web UI with admin privileges.
13. Configuration Man- Prior to license deletion make sure to be in an allowed profile. Failing to do so
agement may result in errors.
14. HDR, Optical Cables, HDR link up time when using optical cables may take 6 minutes or more (up to 20 min-
Link Up Times utes)
15. JSON When JSON API is overloaded with multiple requests, the request queue can
sometimes get to its maximum capacity.
16. Link-Maintenance A port may hang while Link-Maintenance runs on it and the second port’s link
is toggled.
17. Software Manage- Upgrading to 3.6.80xx fails when the command “web http redirect” is enabled.
ment
18. User Accounts If AAA authorization order policy is configured to remote-only, then when
upgrading to 3.4.3002 or later from an older version, this policy is changed to
remote-first.
Revision 3.8.1206
19. IB Director Systems On director systems with a single mgmt, when S01 (or S02) are being plugged
in to the chassis the modules inserted may remain in “powered-on” state and
not complete their configuration.
20. iblink When running iblinkinfo on QM8700 system, there was an error message in
the output. Updated libraries used by fabric inspector and other link libraries to
support HDR.
21. Network Typology When running ibnetdiscover on QM8700 system, the output was unclear.
Updated libraries used by fabric inspector and other link libraries to support
HDR.
Revision 3.8.1174
22. Chassis HA On CS8500 switch system, the switch may perform an unwanted takeover
during system initialization.
23. InfiniBand Switching CLI hangs when running “show fabric sm” repeatedly.
Revision 3.8.1054
1. Chassis Management On director switch systems, the color of the rear fans’ status LED does not
change after removing spine fans.
Revision 3.8.1000
2. Chassis Management False alert of “Insufficient number of working fans in the system”.
3. Chassis Management Minimal fan speed configuration changes periodically.
4. Hostname Sometimes, the DNS server is unavailable when syslog is configured via host-
name. A process might not respond while trying to print.
5. JSON When sending a JSON request that results in a very large output that takes up
all the disk space assigned for the JSON API (10 MB), new JSON requests are
dropped for 1 minute.
6. JSON When sending JSON GET requests with a payload, it may result in the request
being ignored.
7. Linux Dockers When running “configuration text apply” (with “docker no shutdown”), a con-
tainer that is configured as init, may run immediately (instead of waiting for
next boot).
8. PTP PTP announces an interval which is not present for reloading.
9. SNMP In InfiniBand, snmpwalk packets are discarded due to low CPU rate limiters.
10. SNMP On director switch systems, only 1 existing SNMP user is supported with
chassis HA.
11. SNMP SNMP entPhysicalTable and entPhySensorTable are missing DDMI transceiv-
ers power sensors information.
12. What Just Happened WJH engine hangs during parsing of IPv6 packets with a fragment header and
an additional extension header.