On Mon, Feb 25, 2002 at 05:14:29PM -0800, Alvin Oga wrote: > hi ya petro > good you;ve got feedback.. > > > - raw rpm speed by itself doesnt matter ... > > > - 7200rpm ide disks runs hotter than 5400 rpm ide disks :-) > > Oh yeah they do, but fans are cheap, and (for my application) noise > > is irrelevant. If the machines are running too hot, I yell at > > facilities to pump more cold air into the cage. > hey.. thats cheating :-)
No, it's getting my job done. When you've got 5 racks of 2CPU 2u boxes racked to the point where the bottom of one effectively acts as the top of the one below it (100 machines in 5 racks, you do the math), you need lots of fans, and you need lots of cool. There is a 15 degree tempature difference between the front of our cage and the back. (oh, and I've got 10 racks, some of them aren't quite that dense tho'.). > > > - ata-33 ( 33MB/sec) vs scsi-3 (20MB/sec ) comparason doesnt matter ?? > > > - its comparing different "numbers" ... > > > ( but actual data transfer of the same test program is a > > It also matters what kinds of transfers you are doing. Streaming a > > 2 GB media file into memory (for editing) or out onto the network > > is a lot different that making 2GB of changes to a 130GB database. > yuppers.. streaming servers is way different than db servers See petro install test hardware configuration. See configuration put into production. See io subsystem fall over and die. Die subsystem, Die. > > > - transfer speeds are comparable ??? > > In the real world? Probably. YAMV. > yup...and the benchmarks can be tailored to suit ones needs As above, our "benchmark" is to throw the freaking thing into production and watch it die. There just isn't a better test of man or machine than live fire. > > have been in production for 18 months to 2 years (sample size > > roughly 200). None of the 34G IBM SCSIs (sample size 20) have failed > > yet. > ouch.... 40GB IBMers ( deskstar series ) has about a 2% failure rate > for us.... we donno why people still keep insisting IBM ide disks... :-) I forgot to mention another set--limited sample (10 disks) but the IBM 20GB deskstars work fine. We put a rather strenuous load on them (4 drives in a sw stripe) and they are running fine. The load blew up the 3ware controller every day or two, it was taking out the 75gig GXPs like no body's business, but now the 20 gig drives are humming along attached to the ata33 controllers on the MB, and a 39 dollar ATA100 card. > thats across several hundred of um.... 60/80GB deskstars seems lots > better... > - seagates, maxtors, quantums, fujitsu... would be better IDE choices?? > - no failures on them... so far... also over a 2 year sample > period Several years ago I had really bad luck with Fugitsu, 200% failure rate in 6 hours. Bought a drive, plugged it in. Ran for 2 hours then blew up. Took it back to the store, brought replacement home. DOA. The third drive sounded like crap and ran for about 6 months before giving up. > not many people buying them 120GB/160GB disks.... 6 or 8 at a time in a 1U > server... Well, if I could figure out how to get 5 drives in a 1u box, with a good power supply and motherboard, I would. Anything that let's us shrink a monthly recurring cost (figure roughly 750 to 1k per rack per month) shrinking from 10 racks to 5 would save us (using ball park figures, I don't know the details of the finances) 4k a month. For a small startup, that can be the difference between red and black ink. Which is really what the debate about IDE v.s. SCSI comes back to. Yes, SCSI drives are almost certainly better in terms of build quality and speed than IDE disks, but what is the price/performance ratio? Just about everything comes down to economics, and when your talking about business, the "just about" goes away. > -- its fun to ship P3-1.0G without cpu fan.... we take them puppies off > of the heatsink... and it runs cooler with just our itty-bitty side > fans > - guess blowing air straight down has no benefits??? It should, but those CPU fans are designed for "any" case, but if you've got external air blowing straight across the processor, that's going to work better. -- Share and Enjoy.