Which EqualLogic box to buy: 7,200 rpm or 10k rpm?

[cmsJustin]

We run Exchange on 10k drives right now. We are buying an EqualLogic SAN as we move to a virtual infrastructure. We have less than 1 TB of data right now, so I picked 3 TB as the minimum for 5-year growth potential (we are starting to use more for archiving).

Here are the 3 boxes we are considering:

PS300E: 5 TB usable, 7,200 rpm, $56,000

PS3600X: 3.6 TB usable, 10,000 rpm, $59,000

PS3900XV: 3.6 TB usable, 15,000, $67,000

We had our sights set on the PS3900XV originally. EqualLogic came in recently and gave us a demo of the product. After talking with them, they suggested we purchase a PS300E. I told them that we run Exchange on 10k drives right now, are they sure that 7,200 will be fast enough. They said yes, and they have plenty of clients our size (100 users) using the 7,200. We don't need the extra 1.4TB of space, and the price is so relatively close...

So maybe we don't need the PS3900XV. But shouldn't we at least get the PS3600X?

Anyone else using a 7,200 rpm EqualLogic box in RAID 50 with VMware? 10k RAID 50? Both? Any suggestions are appreciated. I feel dirty putting 7,200 rpm drives in my rack...

[joergriether]

hard to believe, i know, but equallogic rocks even with sata drives. I tested the 300 and the 3900 and have to admit the 300 is pretty fast even there are 7200K Drives inside.

best regards

Joerg

[cmsJustin]

How close were they? Within 10-15%?

[5474]

We just got 3 300 and 1 3900. I'll probably added them to the same pool and allow the array to figure out which data should be on the fast drives and let it do the migration. The other option is the 3900 is RAID10 and save that for SQL and Exchange Logs (2000 users).

We've got some time before it has to be in production so I plan on doing a bunch of testing.

[joergriether · Accepted solution]

During our tests we didn´t recognize any diffrence with speed and io´s AT ALL (which made me a little bit confused) - maybe they got a very performant internal algorithm using every last pitch of performance.

[Yps]

We got 5 ESX hosts (14 CPUs) and one fileserver and one Virtual Server connected to our PS400E. We have about 90 VMs running on it, and it works very well.

EQL loadbalance between all disks in the array.

/Magnus

[christianZ]

Check this for comparison:

http://www.vmware.com/community/thread.jspa?threadID=73745

I would get the 3600X. The difference between the 3600X and 3900 is not big, IMHO.

[cmsJustin]

Thanks guys. I think we're going to go with the SATA drives based on your feedback.

I also posted my experience here: http://cmsjustin.blogspot.com/

Thanks for all your help! Especially joergriether for doing the head-to-head.

[java1313]

Can you please share your experience when it will be online ?

How many servers do you attach ?

How many VMs and type of VMs (exchange, sql, how large, etc) are you running ?

Thanks,

[cmsJustin]

When all is finished we will have about 15 VMs on 2 HP DL380 G5 servers. They both have 8-cores, and 32 GB of RAM. Exchange is 100 users, about 200GB, and SQL will be around 400GB. I'm trying to post daily on my blog so check there if you want to keep up to date with what we are doing: http://cmsjustin.blogspot.com

[christianZ]

I want to mention one more time - don't esxpect the same iops performance from sata and sas models (although someone wouldn't see any differences). The differences are definitely there and are more than 20%.

[mrgrid]

Looking at Seagate's drive specs., their 15k SAS drives deliver 30% more IOPS than their 10k SAS drives. They don't publish IOPS on 7200 SATA drives, but I know from experience they are about 35-40% less than 15k SAS drives. The published sustained throughput difference between 15k SAS and 7200 SATA is 16%. This means you should see a significant increase in performance for applications like Exchange and SQL Server using 15k SAS drives, but less difference in file sharing or large sequential workloads.

I'm wondering if EqualLogic's 6 year old BCM 1250 controller technology is finally running out of gas. With SAS drives you really need a controller than sits on at least an 8x PCI Express bus with the latest dual or quad core processor technology.

Intel Server Architecture based storage systems are pushing 21.4 GB/s memory bandwidth, 64Gb/s bus bandwidth with 3.7 GHz dual core CPU clocks.

In comparison, EqualLogic's controller has a memory bus bandwidth of 6.5GB/s (probably not an issue), 532MB/s PCI-X bus (could be an issue) and 1 GHz CPUs (definitely and issue)

I know LeftHand, NetApp, EMC and others all use the latest x86 chipsets.

[joergriether]

let´s read this from the beginning to the end and make yourself a picture;)

http://storagemojo.com/2007/02/16/dear-uncle-storagemojo/

[mrgrid]

Interesting. Not sure what this blog chain has to do with disk performance. After reading this I was left wondering how loyal EQ customers would be if they knew they were buying 6 year old technology. Also, why EQ went cold on the blog after Bob F. asked: "if Left Hand has a more scalable solution than Equallogic. Specifically, increased IO as # of nodes increases, node-level fail-over, use of HP and IBM equipment which does away with proprietary trays, and their MPIO load balancing."

Another good related read on this blog is:

http://www.vmware.com/community/message.jspa?messageID=739337

[cmsJustin]

Please stop. Vendor-bashing just turns me off to the vendor you prefer. So far many of my negative views of LeftHand come from people like you and like <a href="http://cmsjustin.blogspot.com/2007/09/equallogic-performance-confusion.html">this commentor</a> and <a href="http://cmsjustin.blogspot.com/2007/09/equallogic-wants-to-sell-us-stuff.html">this one too</a>.

This thread was started to get performance experience on EqualLogic SATA and SAS drives, not EqualLogic vs. LeftHand.

[doubleH]

amen brother!

[mrgrid]

Yes, I agree and apologize.

In all honesty, you will not see much of a differnece between SAS and SATA arrays if the workloads being generated by the applications are not high enouph to keep all the drives busy.

[beckhamk]

What I want to know is why there is a $24k price difference for a PS100 vs PS3600? Now I am not saying the 3600 shouldnt cost more, but come on $24k for basically upgrading from sata to sas drives AND using the same software thats included on the PS100?! Even is you left the cost of the original 250gb drives and added another $12,800 would be reasonable.

[mrgrid]

The sad truth of the matter is the price of disk drives. For example, if you go to CDW’s web site a 300GB SAS drive is $900 and a 500GB SATA drive is $122.

Based on this e-mail chain it looks like Justin is looking at 100 Exchange users and an SQL Server workload of some sort. If you assume the number of SQL Server users is also 100, and 1 User ~ to 1 IOP (Microsoft Exchange 2003 guide), then he needs a storage solution that can deliver at least 200 IOPS. If you double that to handle peak periods and some file and print you’re looking at a 400 IOPS storage solution.

With this small of a workload you may want to look at starting with smaller nodes. For example (couldn’t resist – sorry) LeftHand sells a 4 drive 2TB 1U node/array for $13,500ea. list. Two of these in a cluster would give you 4TB of raw storage and > 400 IOPS of performance (fully replicated RAID 5). If you still have your heart set on SAS, LeftHand sells a 3.6TB 12 drive 2U array for about $34,800, which can handle well over 1000 Exchange users.

[rush222]

I believe if the # of drives are the same, you'd see similar IOPs but look out for latency. Make the I/O file size bigger than the cache of the EQL so that it hits disk. Faster disks will access data quicker so your latency should drop.