lkml.org 
[lkml]   [1998]   [Dec]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Dumb question: Which is "better" SCSI or IDE disks?
Christopher Smith wrote:
>
> On Wed, 9 Dec 1998, Gregory Maxwell wrote:
> > On Tue, 8 Dec 1998, Christopher Smith wrote:
> > > On Tue, 8 Dec 1998, Leonard Zhang System Administrator ISD RVIB wrote:
> > > > RAID-1 is disk mirror. DPT is cache with RAID. RAID will increase the
> > > > disk performance.
> > > Ugh.. FYI, RAID does not always increase disk performance, and in
> > > particular RAID-1 is guarunteed not to increase disk performance by it's
> > What? Join the real world but. Raid-1 doubles read throughput and the
> > slowdown of writes is negligable.
> A correctly behaving RAID-1 system needs to be able to account for the
> fact that given disk will fail, so interleaving reads is not exactly a
> good idea.
>
Actually, mirroring slows small writes considerably, since you have to
wait for the slower of the disks to access the block. Large (streaming,
anyway) writes are negligibly affected because the writing takes more
time than the latency.

> > > very nature (ok, it's conceivable that read performance could be
> > > increased somewhat, but write performance would, if anything, be decreased
> > > somewhat). Again, if the cache is actually playing a role in this
> > > benchmark, then what you are really saying is that IDE is so fast that
> > > it's 4x faster than DPT's cache memory. Think about that for a minute.
> > Come off it, when you do disk benchmarks you use a dataset sufficently
> > large so that the cache doesn't matter. Linux does caching just fine,
> > DPT's cache is somewhat redundant.
> His system wasn't Linux, it was SCO. If you looked at his data set size,
> it was 9 MB and his cache was 32 MB. So yes, his benchmark was completely
> inappropriate which is what everyone is trying to point out here.
>
> > Almost all other raid levels perform worse (like raid 5 where the fastest
> > write takes 2 reads 1 block XOR and 1 write).
> You are talking about latency, not thoroughput. Read up on RAID, RAID-0 is
> not the only version of RAID with performance advantages.
>
> > The fastest raid array is a combo of raid-0(striping) and raid-1.
> Why would this be faster than just RAID-0?

Because you can read only from the disk that can reach the data first.
Again, small reads are faster, large ones are essentially equal.
> --Chris
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.rutgers.edu
> Please read the FAQ at http://www.tux.org/lkml/

--

Lawrence
~
------------------------------------------------------------------------
Lawrence MacIntyre Oak Ridge National Laboratory 423.574.8696
lpz@ornl.gov http://www.epm.ornl.gov/~lpz lpz@nautique.cind.ornl.gov

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:46    [W:0.117 / U:0.256 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site