On Thu, Mar 06, 2008 at 01:41:43PM -0500, RS wrote: > looking at either a couple of Samsung 750GB spinpoint's or the 1TB Seagate > Barracuda.
Only based on my personal experience, I keep boxes around forever (or at least until gcc stops supporting them) so I keep drives until they die of old age. I've yet to have a Seagate/Quantum drive fail whereas the other brands have. Now, they were all over 5 years old so it doesn't mean much, but a 15 year-old Quantum IDE drive?? The other thing to consider is the duty-cycle of the box. Is it to be left on 24/7, 7/5, etc? How much of that time will it be actively used and at what intensity? What throughput are you wanting to serve and how many simultaneous requests? As in, is SATA what you want or do you want SCSI/SAS? Remember that you can plunk SATA drives onto SAS controllers to start and upgrade to SAS later. I would assume that the manufactures still use the SATA/SAS divide for the quality-control devide when they make drives like they used to with IDE/SCSI but I may be wrong. As has been suggested, diff rsync and raid. Depending on what you are serving, it may be beneficial to have the OS on a separate spindle. If you're in the midst of serving a data stream you may not want the drive to have to seek to run something for the OS, or to fetch swap. When planning the box, remember the memory requirements for fscking a filesystem (I think I remember 1 MB/GB) so that it doesn't hit swap. If you're talking about one TB then that suggests you need 1 GB ram minimum. Upon which will you be backing your data (isn't English wonderful). What will you be using for backup for the 1TB of data? Remember, raid only protects against some drive failure modes, not controller failure, PSU, MB, disaster, etc. Enjoy. Doug.
