On Sat, 03 Apr 2004 21:28:16 -0600 "Christopher L. Everett" <ceverett@ceverett.com> wrote: > > I just read an email of the Linux kernel list saying that Linux software > > RAID > kicks the ass of most hardware raid solutions. Do you have a pointer to that email? Does that email include benchmarking numbers, or a pointer to sam? I find it hard to believe that using the kernel's RAID features blows past "most hardware RAID solutions." It *is* the case that Linux software RAID provides better performance than cheap ATA-based RAID cards like those from Promise and Highpoint (which sometimes are included on motherboards these days). However, that's because those cards are not really hardware RAID, despite the confusing terminology they use. They're really software RAID solutions, and the processing for them has to be done by the kernel anyway. But if someone has performance numbers indicating that Linux software RAID consistently outperforms real hardware RAID solutions, including for a db server under load, I'd really be interested in seeing that. >>Regarding RAID, I believe 0+1 gives the best performance, but it >>requires 4 disks. RAID 5 and 0 also will give a performance boost, and >>require at least 3 and 2 disks, respectively. > > IIRC, RAID 1 gives the best performance under all conditions :), though > it's not terribly safe. Hence the use of 0+1, intended to give you safety + fast performance, without (if you're using software RAID) the computational overhead of RAID 5. > So here asre some of the lower cost options I'm considering: > > SCSI: a multichannel (3 or 4) U160 or better host adapter and several > lightly > used 36GB U160 or better drives, software raid 5. > > SATA: a 3Ware SATA controller, and several SATA drives in a hardware > RAID 5 configuration. > > I suspect the first option would run a touch faster. If you get 4 disks, and you're thinking software RAID, I suggest benchmarking your system under typical usage to see whether a RAID 5 or a RAID 0+1 solution makes more sense for you. RAID 5 is not always a win because of additional computational overhead; it depends on what your usage is. -c -- Chris Metzler cmetzler@speakeasy.snip-me.net (remove "snip-me." to email) "As a child I understood how to give; I have forgotten this grace since I have become civilized." - Chief Luther Standing Bear
Attachment:
pgpp4D0Wsygr7.pgp
Description: PGP signature