JavaScript is required to for searching.
Skip Navigation Links
Exit Print View
Sun Network QDR InfiniBand Gateway Switch

Product Notes for Firmware Version 2.0

search filter icon
search icon

Document Information

Using This Documentation

1.  Sun InfiniBand QDR Gateway Switch Product Notes

Known Problems

General Information and Issues

Reserved VLAN Identifiers

Jumbo Frames

Subnet Manager

Snapshot Dataset Information

Software Information and Issues

Main Board, Management Controller, and Chassis Serial Numbers

High Availability in Partitions

Email Alert Rules

Time Zone Support

Firmware Update Considerations

Host Software and Firmware

Symbol Errors on Internal Ports

Commands Available to Restricted Linux Shells

Hardware Information and Issues

Ethernet Support

Number of LAGs

Unusable Ports

Documentation Information and Issues

smpartition Command Guidelines

VNIC Configuration on Linux Hosts in Gateway Manual Mode

Corrections to Creating Virtual I/O Adapters (Oracle Solaris)

Set Up VIOAs (Oracle Solaris)

Configure IP Addresses on the Data Link (Oracle Solaris)

SNMP V3 Protocol Passwords

Features and Functionality Documented

Acquire the BXOFED Software

Acquire the ConnectX-2 Firmware

Upgrading the Gateway Firmware

Acquire the Gateway Firmware Package (CLI)

Upgrade the Gateway Firmware (CLI)

Known Problems

The following table describes known problems with the gateway.

Setting an alert rule to ipmipet sometimes does not work.
There is no impact to the InfiniBand fabric.

Workaround: After setting the alert rule to ipmipet, set the level to disable, and then set the level to the desired value.

BridgeX reset maximizes symbol error counter.
The setgwinstance, setgwsl, and setgwsystemname commands reset the BridgeX chip, causing symbol errors to maximize.

Workaround: Follow the procedure described in Symbol Errors on Internal Ports.

IP addresses within the Subnet Manager nodes file does not match those assigned by DHCP.
By default, the DHCP server is not assured to assign a consistent IP address to a Subnet Manager node upon reboot.

Workaround: Manually configure the DHCP server to assign a specific and unique IP address to each Subnet Manager node.

No way to set Fabric Monitor console timeout.
The Fabric Monitor will timeout after 15 minutes of idle time.

Workaround: You must login to the web interface and launch the Fabric Monitor again.

BridgeX ports indicate random errors in Fabric Monitor.
Powering on the gateway randomly displays these errors at BridgeX ports BX_0 and BX_1:
  • Peer RcvErrors

  • Peer VL15Dropped

  • XmtDiscards

There is no workaround at this time.

Fabric Monitor does not support Internet Explorer.
Though the Oracle ILOM web interface supports Internet Explorer, the Fabric Monitor does not.

Workaround: Use Mozilla Firefox 3.0 or 3.6.

Oracle ILOM web interface not compatible with Mozilla Firefox 6.0.
Using the Mozilla Firefox 6.0 browser, only navigational control is possible with the Oracle ILOM web interface. No informational output is displayed.

Workaround: Use Mozilla Firefox 3.0 or 3.6.

smpartition start command gives unclear message if the Subnet Manager is disabled.
If the Subnet Manager is disabled before starting a partition configuration session using the smpartition start command, confusing messages are displayed upon issuing the command.

Workaround: Ensure that the Subnet Manager is enabled before starting a partition configuration session.

createvnic command should handle GUIDs with 0x prefix correctly.
GUIDs can be displayed as 0021280001a1623a, 00:21:28:00:01:a1:62:3a, or 0x0021280001a1623a. The latter version is not recognized by the createvnic command.

Workaround: Truncate the 0x prefix of the GUID for the createvnic command.

createvlan command does not support both hex and decimal P_Key value input.
The construct of the P_Key and its membership bit requires the value to be in hexadecimal. Providing and displaying the P_Key in decimal would complicate identifying the state of the membership bit.

There is no workaround.

--clear option of setgwethport command does not restore default parameters.
From the command usage, the --clear option of the setgwethport command should restore to default parameters. This situation does not happen.

Workaround: Use the -clear option.

checkguidfilesftree is displayed with the help all command, but is not supported.
Using the help all command within the Oracle ILOM /SYS/Fabric_Mgmt target displays the checkguidfilesftree command. This command has been deprecated and is not available.
fdconfig command is unavailable from the /SYS/Fabric_Mgmt target.
The fdconfig command is required for configuring the FabricMIB and should be available from the Oracle ILOM /SYS/Fabric_Mgmt target. The command is not available.

Workaround: Log in as the root user, where the fdconfig command is available.

Need tool to display InfiniBand topology with physical InfiniBand entities.
At present, no command provides the relationship of InfiniBand fabric GUIDs, LIDs, and IB ports to each other in a simplified manner.

Workaround: Use the ibnetdiscover command.

Usage of deletevnic command is wrong.
The usage output of the deletevnic command indicates that connectors 0A-ETH and 1A-ETH are supported. They are not.
Stopping or starting the envd daemon initiates a segmentation fault for the FabricMIB.
Stopping or starting the environment daemon on a gateway running as a fabric manager or monitored by a fabric manager changes the FabricMIB configuration due to the new status of that fabric element. Should an snmpwalk of the FabricMIB be issued at the same time the FabricMIB library reloads the updated configuration, internal file pointers might be mishandled, initiating a segmentation fault.

There is no workaround at this time.

Fabric Monitor might fail to load.
During a firmware upgrade or downgrade, certain symbolic links might not be created or become lost. Consequently, the Fabric Monitor will fail to load when requested.

Workaround: Reperform the upgrade or downgrade, using the -force option.

BridgeX manager dies with a base port id collision detected error message.
When two or more gateways within the InfiniBand fabric have adjacent gateway instance numbers (for example 16 and 17), and the gateway with the lower instance number (16) has more than one LAG enabled, there is a port ID collision between the two gateways.

Workaround: Configure all gateways to have even instance numbers (2, 4, 6, ...) with the setgwinstance command.

Gateway hard hangs after reboot.
On rare occurrence when rebooting the gateway, the gateway might experience a hard hang. The last entry in the system log is:

kernel: mlx_bx_core 0000:05:00.0: go bit not cleared

Workaround: Power cycle the gateway by removing both power cords from the power supplies, and then reattaching them.