Hallo Zlatko,
both the originator and me wrote about disastreous performance
with stripping raid-10.
You write about read-before write penalty with raid-5.
I do not understand?
regards
Juraj
> -----Original Message-----
> From: Zlatko Krastev [mailto:acit AT ATTGLOBAL DOT NET]
> Sent: Monday, August 19, 2002 2:37 AM
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: Re: diskpool performance
>
>
> It depends on how the RAID controller implemented RAID-10 (or
> RAID-1E). If
> your write request is small (and random as in TSM DB) you can
> easily hit
> read-before-write penalty usually found in RAID-5.
>
> Zlatko Krastev
> IT Consultant
>
>
>
>
> Please respond to "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
> Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
> To: ADSM-L AT VM.MARIST DOT EDU
> cc:
>
> Subject: Re: diskpool performance
>
> Hey,
>
> what the hell was your raid-10 like?
> Raid10 is by no means slower than raid1, if correctly
> implemented (by the
> means of the raid controller).
> I had it previously on an ICP-Vortex controller, and it was
> real blessing.
>
> But there are another curiosities as well - I see
> worse performance on IBM Raid Controllers using Raid-1E (raid
> 1 consisting
> of more than 2 disks)
> compared to Raid-1.
> More spindles, less performance, .. makes no sense as well..
>
>
> regards
> Juraj Salak
>
> -----Original Message-----
> From: Roger Deschner [mailto:rogerd AT UIC DOT EDU]
> Sent: Wednesday, July 31, 2002 3:54 AM
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: Re: diskpool performance
>
>
> The disk storage pools need to be RAID-1+. It's "other people's data"
> and you are responsible for it, from the moment it is backed up from
> their client systems. As you have discovered, performance is not good
> with anything other than RAID-1.
>
> The Database and the Log also need to be RAID-1. (Mirrored by
> TSM or the
> OS or something) I tried striping (RAID-10, in effect) my TSM
> database,
> and caused a performance disaster that I had to back out of fast. The
> Log might survive RAID-5 without any performance penalty; I haven't
> tried it. Just remember that the Log is strictly write-only until
> disaster recovery becomes necessary.
>
> Roger Deschner University of Illinois at Chicago
> rogerd AT uic DOT edu
>
>
> On Sun, 28 Jul 2002, Seay, Paul wrote:
>
> >There is one caveat Mark.
> >
> >If the loss of a backup is a critical failure to an
> application, then you
> >must Raid-1 or Raid-5 the pool. I have applications that
> take processing
> >cycle backups several times throughout their processing
> cycle. I also
> have
> >servers that have to have the same backup cycle. So, an unprotected
> >diskpool is not a black or white answer to solving a
> performance problem.
> >
> >However, you are correct, RAID-5 is going to perform like
> crap for disk
> >pools unless you have ESS disk which have a very large disk
> cache on the
> >front and the RAID-5 turns into RAID-3 (no reads before
> writes because it
> is
> >always a full stripe write).
> >
> >Paul D. Seay, Jr.
> >Technical Specialist
> >Naptheon Inc.
> >757-688-8180
> >
> >
> >-----Original Message-----
> >From: Mark Stapleton [mailto:stapleto AT BERBEE DOT COM]
> >Sent: Sunday, July 28, 2002 1:46 AM
> >To: ADSM-L AT VM.MARIST DOT EDU
> >Subject: Re: diskpool performance
> >
> >
> >From: ADSM: Dist Stor Manager
[mailto:ADSM-L AT VM.MARIST DOT EDU]On Behalf Of
Dirk
>Kastens
>> Seemed to be a faulty raid controller. We always used raid5 for TSM
>> volumes and never got any errors or bottlenecks. Of course, we use
>> different raids for stgpools and the database.
>
>A question: why are RAIDing your TSM diskpool? There's no need for
>redundancy in the diskpool, since the diskpool is not the
mission-critical
>component of your backup system. It's the *tape* pool you need redundancy
>for, and that's what a copy pool is for.
>
>You *might* RAID 0 your diskpool, so that you stripe across the pool
disks
>for a larger number of read/write heads, but there's nothing but
unnecessary
>overhead when you RAID 5 it. The diskpool dies? You fix the pool, and you
>back up the files again the next night.
>
>--
>Mark Stapleton (stapleton AT berbee DOT com)
>Certified TSM consultant
>Certified AIX system engineer
>MCSE
>
|
