>From what you have told me, I would say that SCSI is definitely the way to go for a >number of reasons.
1) Reliability. SCSI disks tend to be built to a better standard. Most SCSI disks come with either a 4 or 5 year warranty for a reason. They last forever ( have got 8 year old Seagate SCSI disks that are still going today). Most IDE disks are coming with a 1 year warranty (The high end versions are coming with 3). 2) Uptime. I have yet to see an Native IDE system that allowed reliable hot swapping of drives. SATA may offer this, but I wouldn't rely on it for a system that you need to last for 3-5 years. It is still to new of a system. 3) Complexity. IDE busses allow for two drives per channel. Ultra320 allows for 14 (15-1 for the controller card). Fewer cables, fewer cards, less of a mess to wire. If you really want to save money you can go with IDE over SCSI. I have no personal experience with this but from what I have read, they are OK. They are considerably slower than native SCSI, but they seem to work. There sure are enough people selling them. 4) Performance. Ultra320 drives are expensive but they have excellent transfer rates and access times. As for the drives I would recommend either the Seagate Cheetah or the Maxtor Atlas. Both run fairly cool and are very, very, very reliable. You are after VERY high performance go with the 15k drives, but you will new a bunch of them. I think the largest 15k drive that is currently available is a 73Gb. They make 147Gb drives but only in the 10k versions. Whatever you do make sure you go with 64bit controllers, hardware raid, lots of cache memory, and make sure you have redundant, hotswapable everything. If you can't hotswap it, it WILL fail. Hope that helps. Garl > > > On Wednesday, July 23, 2003, at 03:33 PM, Grigsby, Garl wrote: > > > I guess there are a few questions I would ask before I answered. > > > > 1) What is your budget? > < $10k USD > > > 2) How many users? > approximately 5000 > > > 3) What OS/Platform are you going to host this on? > > freebsd > > > 4) What are your uptime requirements? > > Maximal, this a core service, and needs to be bulletproof. > I would be extremely happy if we didn't have to reboot it for > anything > less > than a kernel upgrade. > > > 5) What time frame are you looking at? > > between 3-6 months for fielding it, with an effective service life of > 3-5 years > (if we outgrow it in 3 years we'll be happy ;-) > > > > > > Garl > > > >> -----Original Message----- > >> From: Larry Price [mailto:[EMAIL PROTECTED] > >> Sent: Wednesday, July 23, 2003 3:26 PM > >> To: [EMAIL PROTECTED] > >> Subject: [eug-lug]SCSI vs. ATA vs. IDE raid for >1 TB storage array > >> > >> > >> We have an argument going here in the office regarding > >> how to spec a storage array for a high volume, low-latency > production > >> server > >> that would need to hold and serve approximately 1-3 TB of maildirs > >> > >> (lots of small files being sought by people who get cranky > when they > >> are noticeably slow > >> in arriving) > >> > >> the basic argument is about ata-raid vs. scsi > >> > >> It looks like there are a __lot__ of people trying to sell ata-raid > >> right now, > >> the question is would that work for our application > >> > >> Figured I would ask the knowledgeable and opinionated ( > I'll let you > >> know when I find the list of knowledgable people ;)_ > >> > >> _______________________________________________ > >> EuG-LUG mailing list > >> [EMAIL PROTECTED] > >> http://mailman.efn.org/cgi-bin/listinfo/eug-lug > >> > > > > _______________________________________________ > > EuG-LUG mailing list > > [EMAIL PROTECTED] > > http://mailman.efn.org/cgi-bin/listinfo/eug-lug > > > > > This is a Signature: Someday soon it will have clever sayings > and URL's > > _______________________________________________ > EuG-LUG mailing list > [EMAIL PROTECTED] > http://mailman.efn.org/cgi-bin/listinfo/eug-lug > _______________________________________________ EuG-LUG mailing list [EMAIL PROTECTED] http://mailman.efn.org/cgi-bin/listinfo/eug-lug
