Top 10 Exchange Storage Myths

re: Top 10 Exchange Storage Myths

Joel Christner — Fri, 25 Jun 2010 18:39:45 GMT

Full disclosure: I work for a vendor providing storage solutions for Exchange.

This is a great article and interesting dialog. Our customer discussions have led us to the conclusion that they are concerned that deploying Exchange 2010 at scale does require a tremendous increase in storage capacity, primarily due to the number of isolated copies behind each server (driven by DAG) and the need to increase mailbox quotas (to satisfy user's hunger for email capacity). The over-the-top model with DAG is great from an HA perspective but does create a multiplicative effect on capacity required. We feel that this creates a unique opportunity for customers that want to keep Exchange 2010 on-premises but want a more compelling means of addressing the storage capacity required to support it, and techniques such as hybrid storage solutions, primary storage deduplication, and so on, may provide an attractive option without compromising performance or data protection posture.

If anyone is interested, I encourage you to check out what we've built for Exchange 2010 at: http://www.storsimple.com/microsoft-exchange/

Thanks,
Joel @StorSimpleInc

re: Top 10 Exchange Storage Myths

Bob — Fri, 30 Apr 2010 18:14:59 GMT

I think John is spot on here with some great examples.

AND, when we link to blogs, we need to keep in mind where the blogger works. Mark Arnold does work for a major storage vendor, and of course storage vendors are going to tell you that JBOD is not a cheaper solution.

Also keep in mind that Microsoft doesn't make any more or less money based on your storage solution. Almost all of their money comes from those exensive CALs. So they are really trying to find the best cost solution for you so you will like Exchange more and be more inclined to stay with that rather than switch to a competing email solution. If Exchange is better, faster and CHEAPER, then MSFT FTW.

re: Top 10 Exchange Storage Myths

John — Fri, 30 Apr 2010 17:43:33 GMT

Andrew,

Those blog articles fail to mention a few key points (not to mention the byline of the blog is "rarely factual" which I personally found amusing):

1. a single server vs. multiple servers = multiple servers with multiple isolated copies remove single points of failure.
2. Storage fails and the number one cause? It's not the technology, it's people. You have to buy at least two storage devices to mitigate that (and you need a solution to replicate data).
3. No cost points were provided. Thus the claims are just an opinion.

So here's an example that I recently saw from one of my customers for their E2010 design:

100,000 mailboxes, all 100 message profile. 500MB quotas (though if they go with 1TB spindles obviously they could increase the mailbox sizes without significantly changing the storage numbers). The design was for 3 HA copies.

The SAN storage projected cost (includes SAN licensing and storage) with 1TB 7.2K SATA (RAID-10) was $10 million ($16 million if you went with RAID-5 as it required more disks). This was for 2700 disks. This was the discounted price I should add.

HP sells a 70 disk 5U storage chassis, the MDS-600. Fully populated with 1TB SATA disks, that enclosure goes for $50,500 MSRP. If you keep the RAID architecture it's still the same number disks, so that means you need 39 chassis. That's $2 million for the storage.

If you change the architecture and go without RAID for those 3 copies and deploy on JBOD you cut your storage requirements nearly in half which cuts your cost in half. But heck, let's say to mitigate the move to JBOD you increase your copy count to 4, that's still less disks than the RAID/3copy solution (around 2200 disks) and only around $1.6 million.

And that's only the capital costs. Since both solutions are leveraging SATA, the power costs are roughly the same (the chassis/storage controllers add some additional cost there). Floor costs may be different (depending on rack/space charges) and that obviously has to be factored into the equation (but does that really cost anywhere close to $8 million?).

Ultimately you need to evaluate all the options. Does the model change with more copies and JBOD? Yes. It does mean failovers when disks fails, which may or may not be cool for you or your customers, but that should at least be weighed against the storage options and their associated costs.

For some of my customers they are considering different storage paradigms and more copies to achieve their business goals while reducing costs. Others will leverage existing investments in storage they have. But we should at least help them have an open mind so that they can see where there are potential cost savings.

John

re: Top 10 Exchange Storage Myths

Andrew — Wed, 28 Apr 2010 14:57:25 GMT

How does Microsoft mitigate the increased number of disks needed for a DAS solution?

The (potential) problem is highlighted here:
http://markarnold.blogspot.com/2009/08/exchange-2007-and-das-truth.html

and here:
http://markarnold.blogspot.com/2009/09/how-many-disks.html

While a SAN is not necessary, by the time TCO is factored in it seems like the two solutions, at worst, break even.

re: Top 10 Exchange Storage Myths

Ben — Sat, 03 Apr 2010 01:54:10 GMT

Great article. However, I downloaded the Large Mailbox whitepaper and not even two pages into I found so many grammatical and punctuation errors that it became difficult to consume the actual content. I realize this may sound nit-pickly, but when I read things that are poorly edited it makes me question the accuracy of the actual content of a paper.

re: Top 10 Exchange Storage Myths

orakul — Thu, 01 Apr 2010 23:14:00 GMT

And Microsoft licenses for the additional Exchange servers are free - ha. This is like saying... buy 4 cheaper cars so if the first one breaks down, you have 3 more to drive.

re: Top 10 Exchange Storage Myths

MaximumExchange.ru — Thu, 01 Apr 2010 18:07:30 GMT

Russian version of this post/Русский перевод статьи здесь: http://www.maximumexchange.ru/2010/04/01/top-10-exchange-storage-myths/

re: Top 10 Exchange Storage Myths

online backup news — Thu, 01 Apr 2010 13:07:44 GMT

I think your site is very interesting.

re: Top 10 Exchange Storage Myths

Exchange — Thu, 01 Apr 2010 04:29:45 GMT

Hey Seth,

In order to deploy your mailbox databases on JBOD, you need to have at least 3 database copies - which means at least 3 mailbox servers. Assuming you are deploying 3 copies, it really comes down to two things - ensuring you do not exceed either the IO characteristics or capacity characteristics of the drive. As an example, if your user profile is 150 messages per day, then you can expect your IO profile to be around .15 per mailbox. With a 7.2k SATA disk you can expect to achieve 55 IOPS, so you are looking at a maximum of 300 mailboxes (some fluff factors added). Now you look at your capacity and dial down the number of mailboxes there; so maybe for 2GB mailboxes you can still handle 300 mailboxes per disk. So now you know that you need 2 databases to handle your 500 mailboxes, which means you'll have a total of 2 disks per server.

As for sleep loss - I think the key thing there is that you are building enough redundancy in teh system so you don't have to worry. By having 3 or more copies, if you lose a single copy, you still have two other copies to fall back on - one that is the active copy, and another that is a passive copy and a seeding source for third copy when you replace the disk.

Now as to whether you want to deploy this architecture for your organization, will really depend on a multitude of factors, one of which will be cost.

Hope this helps,

Ross

re: Top 10 Exchange Storage Myths

Seth — Thu, 01 Apr 2010 00:15:05 GMT

Using SATA JBOD disks blows my mind to a point where I doubt I could sleep at night. Do you have a visio or such of how you would utilize this scenario for a mid sized company of say 500 mailboxes? I looked at the Excel download, but it isn't helping me visually get the concept.