Brock Wolfe wrote:

I am looking at building a high reliability, fault tolerant file server to store 300 plus GB worth of files. Reliability is needed b/c a sysadmin will not always be around/on-site to ensure full backups are made regularly. I am assuming RAID 5 array, for data and seperate drive(s) for OS (RH 7.x), redundant power supplies. Any suggestions (voice of experience) on:

• Motherboard/CPU combo's
• Server Cases
• RAID host cards (3ware-IDE, Mylex-SCSI)
• High(er) MTBF IDE, SCSI Hard drives (I don't want another 60/75GXP fiasco) <g>.
• Tape drives (DLT, DAT, etc)

When I worked in Calgary, we had clients that we visited very rarely and had very low remote sys admin duty requirements. Everything was automated and the users did perform the regular tape changes etc as required. The onus can be placed into a group of competent individuals to perform what is really a mundane task. Shawn commented earlier that if you are going to span multiple tapes, that you will most likely want to have a tape library solution ... more expensive, but your options open up much more as well. I've also seen solutions that a "snapshot" of the data is copied to another set of drives and those drives are backed up to tape ... multiple or not, doesn't matter, the original data is free to carry on it's merry way and the users are able to change the tapes while they're at the site.

You haven't really given any indication as to what kind of data will be on the drives, how often it is updated etc ... could you use incremental backups vs using full backups continually? What is the purpose of the file storage? Are they just huge files (such as what Shawn has to deal with) or are they relational databases? If it's relational, how many transactions are you talking about -- user load etc ... if it's high, or you're performing a fair amount of random write activity stay away from raid5 if you care about performance in the long term ... and don't get tricked into using some kind of write-back caching to compensate for the abysmal raid5 write performance for those types of files ... if the cache fails, your array goes down and your data is hosed -- yes, I've seen it happen and it's not nice. If it's a *file* server, is it's primary duty to serve files to PC's on the network? if so, raid5 is more useful for this type of deployment, also, what about the possibilities of a nas or san ... The muug page describes the october 2001 meeting which discusses the topics of nas/san.

We run about 288GB hot-pluggable, hot spared, across two identical systems in a raid 0+1 configuration, multi-channel scsi controllers on multi pci channels networked with a fiber connection ... everything is redundant right down to the power supply .. UPS's on everything including the network. Not cheap, but works well. Also have (relatively) cheap PC's running linux in various branches across the country which are easily re-creatable from a hardware and software configuration perspective... there's not much redundancy in this except for the easy replaceability.

Dan.