RAID configuration for Thalia production cluster
The Thalia production cluster needed large amounts of disk storage available locally to hold its content. To provide this, a RAID array was built in the cluster, as follows:
Base hardware for the Thalia production cluster in the HP DL380 G5. It has drive bays for 8 SATA or SAS drives. We had it populated with 8 146 Gb dirves, built into 2 RAID arrays:
A RAID 1 array of 2 disks, with a total of 140 Gb, for the OS.
A RAID 5 array of 6 disks, 4 for data, 1 for parity, and 1 hot spare, with a total of 538 Gb, mounted at /home. It holds the picture content files.
If a drive in the RAID 5 array fails, the spare disk will automatically be used as a replacement for the failed drive. Having the parity drive in place means that 2 drives will have to fail before data lose occurs.
The RAID 1 array is performing direct mirroring, so we would have to loose both disks before data is lost.
Before a RAID 5 array can be deployed, the hardware RAID controller in the DL380 G5 needs to be upgraded. The card itself does not change, but it does need to have a memory upgrade. This upgrade is a Flash memory module that acts as a write cache. Writes to a disk are first stored in the write cache, and are only removed after the write has been committed to disk and checked. If the case of a power failure, any write that was in progress at the time of the failure will be automatically rolled back to the previous state, and then rewritten from the cache to disk and confirmed. This prevents data loss or corruption due to power loss in the majority of cases.