Wanda has identified the critical choices;  there is no single, right answer
because the performance factors intersect at different points for nearly
every situation -- if only because most sites are at different points along
the evolutionary development of IT infrastructure, different sized disks for
RAID, different RAID technologies (ESS vs. EMC vs. native/local drives,
etc.), and different performance ranges for a given TSM server box.

And it does matter how you decide to carve up your disk pool resources. Just
as different tape capacities & technology will influence your choices for
how tape pools are configured, variety in the type and capacity AND number
of channels accessing a set of disks will dictate the kind of performance
you can achieve.  So, if you think your TSM server system can handle 25 (50
in today's terms) sessions concurrently, then you measure the disk I/O
performance choices against that level of concurrency.  Most folks (these
days) disregard the effect of distributing many logical (TSM) volumes over
some number of physical volumes;  increasing the number of logical volumes
per physical gets that many concurrent writes queued to the device.  The
more of each, the more sessions can be sustained without waiting for
"mount-point".

I rather NOT use RAID-5 for TSM disk pool volumes... Why pay the penalty for
calc & writing parity when you don't expect to read it more than once?!?
This data is so transient (in general, it exists for less than 48 hours)
it's not worth sacrificing ANY performance -- we really want to get the
backups AND migration done as quickly as possible, right?  These days, I
recommend RAID-0, striping without parity, if any at all;  on most moderate
sized Unix machines, no striping at all, and as many smaller drives as can
be handled -- these days that translates to 36 or 72 GB drives, fill the
drive bays, get as much SCSI separation as possible... do the math on
dividing each physical drive into some 7-10 logical drives.  For RAID
anything, Gianluca's 7 or 8 physical drives per RAID config is a good
high-end count.


Don France
Technical Architect -- Tivoli Certified Consultant
San Jose, Ca
(408) 257-3037
mailto:[EMAIL PROTECTED]

Professional Association of Contract Employees
(P.A.C.E. -- www.pacepros.com)



-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:[EMAIL PROTECTED]]On Behalf Of
Gerald Wichmann
Sent: Tuesday, May 28, 2002 3:30 PM
To: [EMAIL PROTECTED]
Subject: Re: allocating disk volumes on RAID5 array


That's more along the lines of what kind of info I'm digging for - but
doesn't quite address how one goes about coming up with a number or size.
There may not be a "right" or "single" answer however you must admit there
is a generalized "right" answer. Consider for example that while RAID-5
arrays can vary in the # of disks assigned and size of them, there is a
generalized rule of thumb as Cordiali pointed out. Too few or too many disks
in your RAID-5 array and you can have performance implications. There's a
sort of sweet spot for creating RAID-5 arrays and keeping that in mind there
should also be a similar sweet spot in how many volumes one might want to
assign. All in all though I still wonder if it really matters a whole lot..
Whether you have 1 write going on or 10, you're still striping across the
array and I question whether you'd really see much difference one way or
another. Might be an interesting thing to try various variants of..

Regards,

Gerald Wichmann
Senior Systems Development Engineer
Zantaz, Inc.
925.598.3099 (w)

-----Original Message-----
From: Prather, Wanda [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, May 28, 2002 2:38 PM
To: [EMAIL PROTECTED]
Subject: Re: allocating disk volumes on RAID5 array

Well yes, it does matter.
Trouble is, there is no "right" answer.
It's a performance thing.

Assuming you have multiple clients backing up concurrently to your disk
pool, TSM will start as many I/Os as there are clients sending data, up to
the number of  TSM "volumes" in your disk pool.

If you have more "volumes", then you get more I/O's in flight concurrently.
That's a good thing and will improve performance, until you get "too many"
in flight, then the effect of yanking the heads around degrades performance.

It's even harder to figure out what is optimal in a RAID situation, since
you don't have a 1-to-1 correspondendence between your TSM "volumes" and
physical disks.  And most RAID setups have some cache that acts as a buffer,
and that helps improve performance but further disassociates the number of
concurrent writes from the number of physical disks.

So think about it this way:  How many concurrent WRITES do you want to occur
in that RAID pool?  Pick a number, and create that many TSM "volumes".



-----Original Message-----
From: Gerald Wichmann [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, May 28, 2002 5:01 PM
To: [EMAIL PROTECTED]
Subject: Re: allocating disk volumes on RAID5 array


Yes I'm well aware of the different pro's and con's to using various levels
of RAID vs non-RAID. In my application protection is paramount and mirroring
is simply too wasteful to use. I've used RAID5 repeatedly in the past in
regards to TSM and have always been very happy with the results. So the
issue here isn't really what to use or not but rather whether there's any
pro's or con's on the way you go about creating volumes on a RAID5 array.
E.g. lots of smaller volumes or fewer large volumes? Having lots of RAID5
arrays as was also suggested isn't really practical because these days it's
rare you don't have fairly large disks (18 or 36GB each) so in that example
of 100GB you're really only talking 1 RAID5 array of 4-5 disks.

Bottom line is I was just speculating out loud perhaps on whether there were
any pro's or con's to how many volumes and what size one would make the
volumes on a RAID5 array. Say you had a 100GB RAID5 array. Would you create
10 10GB volumes or 2 50GB volumes? Does it matter since it's all just going
into a big array?

Regards,

Gerald Wichmann
Senior Systems Development Engineer
Zantaz, Inc.
925.598.3099 (w)

-----Original Message-----
From: Gianluca Perilli [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, May 28, 2002 1:35 PM
To: [EMAIL PROTECTED]
Subject: Re: allocating disk volumes on RAID5 array

Hi Gerald,

I think you have to consider that if you use RAID5 logical drives, you have
to calculate and write a parity every time you write any data on the disk:
so if you have a write intensive application, RAID5 is not so efficient as
other RAID protections (1,10, etc); if you have instead read-intensive
applications RAID 5 is a good choice because it gives you the possibility
to use more physical drives concurrently.
Furthermore RAID 5 is the most efficient protection regarding the optimal
usage of the available phisical capacity, and it is more and more efficient
as the number of physical drives in the array increase; but at the same
time as the number of physical drives increase, the performance goes down
(because you have to calculate the parity on a larger number of blocks):
probably the best compromise is a number of 7/8 disk drives/array.
I hope this helps.



Cordiali saluti / Best regards

Gianluca Perilli



Gianluca Perilli
Tivoli Customer Support
Via Sciangai n0 53 - 00144 Roma (Italy)
Tel. 06/5966 - 4581
Cell. 335/7840985




                      Gerald Wichmann

                      <gwichman@ZANTAZ.        To:
[EMAIL PROTECTED]

                      COM>                     cc:

                      Sent by: "ADSM:          Subject:  allocating disk
volumes on RAID5 array
                      Dist Stor

                      Manager"

                      <[EMAIL PROTECTED]

                      .EDU>





                      28-05-02 21.14

                      Please respond to

                      "ADSM: Dist Stor

                      Manager"








Since a RAID-5 array shows up as one big filesystem, what's the best
strategy for determining how many and of what size disk pool volumes to
create for your primary disk storage pool? For the most part I don't think
it really matters unlike allocating volumes on individual disks but perhaps
I'm not considering something.

Thanks..

Reply via email to