Cisco shipped UCS servers with rotten RAID settings
This one weird trick will stop your telephony server staggering
If you've been wondering about the server performance in your Cisco Business Edition 6000/7000 telephony system, wonder no more: The Borg has issued a field notice that the system shipped with misconfigured RAID.
The Cisco field notice advises sysadmins that the correct settings for the kit are as follows:
Read Ahead Policy = Always
Write Cache Policy = Write Back
Strip Size = 128KB
The factory configuration error exists in the BE6000M, BE6000H, BE7000M, and BE7000H appliances that shipped with the Cisco BE 6000 and 7000 telephony systems.
If you have an affected system, you've probably noticed something like this: “Installations of Cisco Collaboration applications from ISO (disk image) files directly located in the local datastores might take six to seven times longer than normal. For example, a Cisco Unified Communication Manager installation will take six or more hours versus the typical one hour install time.” (Emphasis added)
Sorry, sysadmins: it's going to be a long night. While the read-ahead and write cache policy changes are just check-box changes, the strip size change requires a rebuild of the RAID logical volume – and all of the backup, virtual machine management, and restore-wrangling that this involves.
Cisco explains the issue arose when it upgraded from Unified Computing System M3 to UCS M4 in the telephony systems. That entailed a change of RAID controller, but engineers forgot to update the installation script. ®