What do you think of this spec of SAN?

[netapp2209]

Looking at a Netapp 2040 with an extra part-populated SAS disc shelf running iSCSI.  There will be 28 x 450GB 15k SAS discs in total.  I want to make sure we get better performance than we have without a SAN and with future capacity.

How many users do you think this could service assuming every user is a heavy exchange user, heavy sql server user and we have 20 servers in total (mixed SQL, Exchange, file and application servers).

The servers are all running on top of ESX vSphere for which the vmdk files will hopefully live on the SAN.

thanks in advance.

[radek_kubka]

Hi & welcome to the forums!

Performance sizing is a dark art, where straight answers doesn't exist!

If you are end-user, then work with your NetApp partner and/or NetApp technical resource to do proper sizing using available tools.

If you are reseller, you can access these tools via Field Portal. Still, if this is your first experience with sizing, I'd strongly recommend working alongside with your local NetApp techie!

Regards,
Radek

[netapp2209]

Oh OK thanks Radek, I've been quoted the above SAN but without me giving any specs of our servers which I find a little odd.

I thought that surely if you are going to spec a SAN, you're going to need to know what is required of our disks currently in terms of IOPS etc...

I'll ask our reseller to use the Netapp sizing tools as you suggested, then send me the output.

[netapp2209]

Our reseller just told me that our installation was too small to use the Netapp capacity planning tool on......what are your thoughts on this please?

[radek_kubka]

Change the reseller!

No, seriously - I am not getting this explanation. At least some simple guidance should be offered, especially when you explicitly ask for it.

Running just MS Exchange sizer is a 5-minutes exercise & this in itself can give the rough idea whether this is the right setup for you. SQL is arguably more complicated, but still doable (by running custom scripts on your SQL servers & then uploading collected statistics into DB sizer)

And BTW - if I got it right, you are looking at 24, not 28 spindles - 12 in the main unit & 12 in the additional shelf (that's the default number for a part-populated, 24-bay SAS shelf)

[lwei]

How many Exchange and SQL users do you need to support? What are the user profiles (IOPS & capacity per user)? Normally, sizing starts from user requirement information.   -Wei

[netapp2209]

Exchange - 220 REALLY heavy users, we use 2003 which I'm aware isn't as efficient as 2007 onwards. Estimated 2 IOPS per person = 440 IOPS

SQL - 150 medium users, no idea on IOPS here.

We did some capacity planning when we had ESX installed and found that the average IOPS across all or systems came to 1600 IOPS.

Does that help at all?

[radek_kubka · Accepted solution]

Hi,

If your total workload oscillates around 1600 IOPS, then I am fairly sure this setup will easily cope with the performance requirement. In fact I reckon you may have about 50% headroom, i.e. you can double the load & the filer still should be fine.

Bear in mind though this is all just a best effort estimate, without in-depth knowledge about your environment, so as they say: your mileage may vary!

Regards,
Radek

[lwei]

I agree, from IOPS estimation it should work. With Exchange 2003 Microsoft and NetApp best practices recommends separate database and logs onto different spindles.   -Wei

[radek_kubka]

Hi Wei,

I actually couldn't find any up-to-date doc describing Exchange 2003 best practices.

This one is from Nov 2006 & actually doesn't recommend separate aggregates for logs:

http://media.netapp.com/documents/tr-3350.pdf

Having said that, personally I would carefully read the TR-3578 from Brad Garvey - although written for Exchange 2007, some recommendations are applicable for 2003 as well:

http://media.netapp.com/documents/tr-3578.pdf

Page 11 gives nice view on single vs. two aggregates dilemma.

Regards,
Radek

[lwei]

Hi Radek,

You are right. TR-3578 should be a good reference for Exchange 2003 as well. Separating database and logs is a "good to do", not "must do".

Regards,

Wei

[netapp2209]

Thank you.

I've just now had a meeting with a Netapp reseller who is 100% confident that a fully loaded 2040 with an additional shelf populated with 16 discs (out of 28) will suffice for our IO quite easily.

Thanks again for the pointers.  Without tweaking Exchange I think we'll benefit from a SAN anyway in that currently we're using local storage for Exchange with only a 512MB cache.  The 2040 comes with 4GB so this will surely help in itsself let alone the fact there are more spindles to go at!

[rgraves2572]

We recently implemented a cluster FAS2050 running 20 x 300 GB SAS along with 14 X 1TB SATA. The 2040 is a more powerful CPU but I would be carefull with the FAS2000 series for heavy workloads.

We found that the FAS2050A could handle the workload of 400 global users for Exchange/SQL/Sharepoint running on Hyper-V, the additional tasks such as SnapVault backups, SnapMirror to DR, and Deduplication over taxed the system.

Of course we were in a global 24 hour enviroment and backups had to be fast low impact on the host. Hence, the use of NetApp snapshots for backups and the need for 1 hour SnapMirror updates to DR.

This is our experience and you really need to engage NetApp Pre-Sales Engineer to gather your requirements and recomend  a solution. If there is a chance you may need a head upgrade due to growth or under estimating load in the future then I would avoid using internal disks. Upgrade with external shelf is much easier then internal disks because a waste once you go to 3000 serios and above.

-Robert