Workaround If Auto Negotiation Fails
sbodily 100000RUAA Comments (2) Visits (6092)
In most cases, configuring eth0 and eth1 as either private or open networks using auto negotiation results in no problems whatsoever. But on occasion, the ability of the Ethernet card to correctly negotiate full duplex with a particular switch or router appears to fail and the adapter comes up in half duplex mode. If the switch really is configured for full duplex, the switch may be reporting a problem (or it may go unnoticed) and network traffic on the HMC will slow. Obviously, this can have negative consequences for the performance of the HMC.
A workaround for this problem is to manually turn off auto negotiation on the adapter and set it to full duplex. The ethtool command on the HMC can be used to execute this two-step workaround. However, the command can only be issued by root. This means the administrator will need to bypass the restricted shell:
1. Login to the HMC command line interface using the hscpe userid and password
2. Obtain the serial number of the HMC using the lshmc -v command (look in the SE field)
3. Call the IBM Hardware Held Desk and provide the HMC serial number in order to obtain a temporary pesh (Product Engineering Shell) password
4. Logged in as hscpe, enter the command pesh
5. When prompted, enter the pesh password suppled by the Help Desk
6. Enter the command, su -
7. Enter the root password when prompted
Once you are logged in as root, you can use the following ethtool command.
1. Enter ethtool ethN for the adapter (N) you want to look at, and you'll see it's present settings
2. Turn off auto negotiation by issuing the command ethtool -s ethN autoneg off
3. Manually set the adapter to full duplex with ethtool -s ethN duplex full
4. Check your work with ethtool ethN
5. DO NOT reboot the HMC.
The setting will be changed to full duplex until the next reboot, at which time the default settings (e.g., auto negotiation) will be restored and the problem may resurface, requiring you to repeat the process.
It's important to note that this does not occur uniformly across the board with all switches or routers. Therefore, this workaround may not be needed. But you should be aware of this workaround, and that this problem is under active investigation by IBM.