2009-07-15 12:14 PM
While it's true that the more disks an Aggregate has, the better performance it gets, but isn't there a point of diminishing returns?
Does NetApp have a chart or document showing the max amount of spindles an Aggregate should have, before adding more disks becomes less effective?
2009-07-16 01:46 AM
I think the actual answer is really as many as you can have.
Bear in mind that there is a 16TB cap on aggregate size, so realistically these days, with large drives, this can limit the spindle count:
There are some additional considerations around (rather extreme) high availability - e.g. it's recommended to put database tables & logs on separate aggregates, but this can be easily done in a clustered setup where you have (at least) two aggregates anyway.
2009-07-16 08:03 AM
Really just more disks = more performance....RAID DP allows you to go a lot larger with RAID Groups than traditional solutions and aggregates allow you to get more disks (i.e. via sets of RAID Groups) into that same overall performance envelope.
2009-07-16 09:13 AM
So, there is no law of diminishing returns? Would a 112 disk aggregate perform better than an 80 disk aggregate? Yes, but my question is: How much better? Doesn't the performance increase based on disk become less and less after a certain amount?
2009-07-17 02:28 AM
Performance is really a complex subject.
First of all, to have any benefit from a 'faster' setup, you need to have a suitable workload(s) to make use of the underlying performance.
In a typical random IOPS environment your (theoretical) performance grows linearly when additional spindles are being added. And yes, there is a cap somewhere - the most obvious one is a storage controller capability, but this is usually addressed by the fact that they support only as many attached disks as it makes sense.
The trick is you may have a bottleneck somewhere else in your environment which will cap you effective performance first: networking, a server itself, etc.
2009-07-21 12:29 PM
Thank you all. So there is a linear performance growth based on the amount of disks added.There may be other bottlenecks (memory, capacity, etc..), but the amount of disks is in direct relation to iops. In the past I heard of a point of diminishing returns, apparently that is false.
2009-07-21 02:29 PM
Performance.. there are soooo many factors involved that its really impossible to give a clear answer as you can hopefully agree to.
I have had customers throw workload on controllers (110 VCB backups at the same time) then watch the CPU increase to 70%
and claim there is a performance issue.. That is how simplistic people can be. They also chose to disregard that CPU dropped when
the backups stopped. If you step on the gas pedal your engine will roar, take your foot off and engine will relax.
My point is that performance is something different to everybody, there does not seem to be a clear definition of a performance problem
everybody can adhere to, sadly enough.
I have heard a lot of funny stuff out there, the funniest one that some competition use is that "WAFL is old"!!
Uh. how old is SMTP and how widely used is that?
2009-07-25 09:36 PM
Sounds like a good summary -- basically yes, more disks = more performance but can't leave out all the other factors that play in (complete agreement with Eric as the complexity around performance testing/validation...much less helping people understand it ).
2012-02-28 07:45 AM
In fact, is not always the best to add more disks, you could install some PAM modules, and it will give you more performance and the disks you had will be less punished. Depends on the number of disks you need to add to gain more performance, PAM modules could be less expensive , also it depends on every situation. Remember the way WAFL works, so if you had less available storage in your array, you will notice then some performance impacts, so be carefull with the available space that you had on your arrays.
Hope it helps.