Technically, doing any of this (mixing drive speeds in RGs or drive sizes in RGs) will work just fine. The only issue is the potential of performance degradation in the future once\if the system is under stress.
The drives spinning slowest will take the longest to have writes flushed to them - thusly forcing the CP to take longer and impacting the entire system. This is true within RGs or across multiple RGs in an aggregate. Heck, this is true even if you have multiple aggregates and one is comprised of slower disks than the others (or is slow due to other reasons such as workload, fragmentation, etc) - a write (CP) is a GLOBAL ENTITY and it is atomic in nature: if you have data in a CP for multiple RGs\TradVols\Aggrs and one takes forever to flush to disk, then the WHOLE CP takes forever.
That said, all of this 'chicken little\sky falling' stuff is just conjecture. On a lightly loaded system, you'll never see issues. Only under stress will all of this stuff bubble to the surface.
Oh - and FWIW, if you write data in a less-than optimal fashion, then subsequent reads will have the same issues (higher latencies), potentially causing performance problems.
Remember - all of this could still take place and you could potentially never see the effects depending on the circumstances and how the applications handles increased latencies.
I hope all of that is clearly written?
Glenn
-----Original Message----- From: maclean@cs.mcgill.ca [mailto:maclean@cs.mcgill.ca] Sent: Thursday, September 21, 2006 2:41 PM To: Glenn Walker Cc: toasters@mathworks.com Subject: RE: Mixed disk sizes within a single aggregate
I always thought you would have problems mixing speeds in a raid group. Specifically you would see 'failing servo motor' autosupport messages.
Or are you suggesting raid groups segregated by speed and only mixing speeds within an aggregate made up of the speed based raid groups?
-Matthew
On Thu, September 21, 2006 11:01 am, Glenn Walker wrote:
Exactly :)
Consistency is key: always try to keep same disk sizes speeds in aggregates. I'd feel more comfortable adding disks with faster speeds than larger sizes - there will be no future performance degradation
from
this.
Glenn
-----Original Message----- From: Crawford, Mark (CBC) [mailto:Mark.Crawford@CapBlueCross.COM] Sent: Thu 9/21/2006 10:53 AM To: David Knight; Glenn Walker Cc: Paul.Brosseau@netapp.com; jeff.mery@ni.com; toasters@mathworks.com Subject: RE: Mixed disk sizes within a single aggregate
When the stripes fill the equivalent of the smaller disks, wafl will start striping only on the larger disks. This will cause performance degradation due to "hot" disks.
-----Original Message----- From: owner-toasters@mathworks.com
[mailto:owner-toasters@mathworks.com]
On Behalf Of David Knight Sent: Thursday, September 21, 2006 9:44 AM To: ggwalker@mindspring.com Cc: Paul.Brosseau@netapp.com; jeff.mery@ni.com; toasters@mathworks.com Subject: Re: Mixed disk sizes within a single aggregate
I think wafl stripes across all disks in an aggregate. I'm not sure how it respond to having one raid group with larger disks
-
likely it will not use part of them. I'm pretty sure netapp recommends
all
disks in an aggregate be the same size, or, it will assume they are
all
the size of the smallest disk. Of course, I could be wrong . . .
David
Remember that WAFL still writes across the entire aggregate. Having
a
slower RAID group in an aggregate of faster raid groups would be akin
to
having a slower disk in a RG of faster disks, would it not?
Glenn
From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Brosseau, Paul Sent: Wednesday, September 20, 2006 8:05 PM To: Jeff Mery; toasters@mathworks.com Subject: RE: Mixed disk sizes within a single aggregate
Mixing disk sizes in an aggregate is not a problem as long as you create RAID groups for each kind of disk. WAFL creates stripes at
the
RAID group level. For best results create complete RAID groups each time you add disks to an aggregate.
Paulb
From: Jeff Mery [mailto:jeff.mery@ni.com] Sent: Wednesday, September 20, 2006 3:30 PM To: toasters@mathworks.com Subject: Mixed disk sizes within a single aggregate
Greetings fellow toasters!
<Background> We're looking at moving our 2 FAS940 systems from tradtional volumes to flexvols + aggregates. </Background>
It would seem to me that the same rules and guidelines for creating traditional volumes now apply directly to the aggregate level (for
the
most part). By rules and guidelines I mean things like trying not to mix disk sizes, try to avoid volumes (now aggregates?) that span FC adapters, etc.
Are any of these things still a concern on modern versions of ONTAP (7+)? Does anyone have any best practices they'd be willing to share in regards to aggregate creation? NOW says "make them as big as
possible
using as many spindles as possible", but that doesn't really
help much.
We use our filers for unstructured data only; cifs + nfs but no databases, no snapmirror, no snapvault, etc..
TIA, Jeff Mery - MCSE, MCP National Instruments
----------------------------------------------------------------------
--
"Allow me to extol the virtues of the Net Fairy, and of all the fantastic dorks that make the nice packets go from here to there.
Amen."
TB - Penny Arcade
----------------------------------------------------------------------
--
Confidentiality Notice: This e-mail is intended only for the personal and confidential use of
the
individual to whom it is addressed and may contain information that is privileged, confidential and protected by law. If you are not the intended recipient, you are hereby notified that any use or disclosure
of
this information is strictly prohibited. If you have received this message in error, please notify the sender immediately by reply e-mail and delete the original message. Your compliance is appreciated.
-
Glenn Walker