Rocks Cluster Installation on 64-bit/Hardware RAID Setup
Last week I finished up an install of Rocks Cluster, a Linux distribution prepackaged and configured for use in a cluster environment. The current version is based on CentOS 4.1 and comes highly recommended, but I had some issues with the initial installation.
First off, the installer fails to copy the needed packages to disk because it segementation faults when, just prior to the copy, it attempts to format your hard drives. The error that you receive only tells you that some of the package header lists cannot be read, so it took a little while to realize what exactly was going wrong. Formatting the drives ahead of time solved this hurdle.
Secondly, once installed on a 64-bit SMP machine with 3ware RAID this CentOS-based distribution envinces some of the same problem I’ve encountered with CentOS proper. The machine kernel panics when trying to boot with an SMP kernel and is highly unstable with the single processor kernel loaded. To resolve this problem a few things have to happen. You need to boot successfully into the single processor kernel and remove, at a minimum, the haldaemon and dmraid packages. The haldaemon causes crashes trying to detect and abstract the hardware in question while dmraid is not necessary as we’re dealing with a hardware, not software, RAID configuration. If stability problems remain it may be worthwhile removing the i386 device-mapper package as well.