EqualLogic SAS vs. SATA

[bburke]

We're looking at purchasing an iSCSI array from EqualLogic to use with our existing ESX 3.0.1 servers. Can anyone that currently uses an EqualLogic array give me some feedback on the performance that they have experienced with the SAS or SATA drives?

[RParker]

SAS is SCSI, you won't outperform SAS drives by using SATA. However, SAS drives are limited to 400g. SATA drives are 1TB or more. So that's the difference.

You want performance or do you want space?

[bburke]

Right. We're looking for the best performance possible. But, if the difference between how the two types of drives handle the VMware storage load is negligible then we'd be open to the SATA drives.

[bburke]

Side note...EqualLogic Sales told me that the SATA drives would handle our VMware needs just fine, but I wanted to see if anyone was actually seeing that in production.

[doubleH]

it depends what you are putting in your virtual environment. i have the typical....DC's, File/Print/Application Servers/Oracle/WSUS/WSS/Symantec AV/etc all in a VI3 environment utilizing EqualLogic PS100E SATA arrays and it's running fine with no issues. I started with a single PS100E and now up to 2 arrays with the 3rd on order.

If you found this or any other post helpful please consider the use of the Helpfull/Correct buttons to award points

[bburke]

Cool. We're looking to do several dozen vm's for our Citrix environment, so the CPU and Memory demand's will be higher most of the servers that you mentioned, outside of possibly Oracle and the Application servers (are they running terminal services or Citrix?), but I don't know how different the storage load would be for a Citrix vm vs. what I would call an infrastructure vm.

With the two PS100E's that you have, are you load balancing across them? Do you have any other servers, other than your ESX servers, using them for storage as well?

[doubleH]

I am also running Citrix Access Essentials (slimed down version of MPS). I read some threads on Citrix and it appears that you don't get as many users per citrix vm compared to a physical box....just as a fyi. Yes I am load balancing my arrays so if i have a 500gb lun the arrays each hold 250gb. Great thing about that is that I am spreading the IO over double the controllers/RAM/Cache of the arrays. I have my backup server attached to the san because i needed the disk space to hold my vRanger backups and also put one of my physical oracle boxes on it because it was running out of space.

Not sure what models you are looking at, but EQL is having a firesale on their 'old gear'..PS100/300/400 because the of new Dell EquaLogic 5000.

If you found this or any other post helpful please consider the use of the Helpfull/Correct buttons to award points

[bburke]

Did you go to the second array for performance reasons or extra storage? BTW, what is the usable capacity of the 100E?

[chucks0]

One other note . . . . since the SATA drives typically have twice the capacity of SAS drives, you may end up putting more load on them than you would SAS drives eventhough they have lower performance. We opted for SAS drives in our setup because I didn't want to regret the decision later. SATA may offer enough performance for what you are doing today, but not down the road.

That being said, our PS300E SATA box held it's own performance wise with our EMC CX500 with Fiber Channel drives.

[doubleH]

2nd array was for extra storage, but the performance benefit of adding more arrays is also awesome. 3rd array is bring purchased because of the awesome pricing right now. Usable space of a PS100 with 250gb drives and RAID 50 is 2tb (remember out of the 14 drives 2 are reserved for online spares). If you have a PS100E and decide you need space you could also buy 500gb drives and swap them out, but you won't get the additional performance benifits that i mentioned before.

If you found this or any other post helpful please consider the use of the Helpfull/Correct buttons to award points

[RParker]

Well it's certainly NOT neglible, and this makes me want to NOT use Equal logic if that's their stance, because that's retarded.

SATA @ 7200 RPM is no where NEAR performance of SAS @ 15K RPM, not to mention burst speeds, and cache, and SCSI backplane.

Maybe they ASSUME it's going to meet your needs, but why then is there a performance difference in the real world? Ask them to prove to you by showing you numbers. running 1 VM on a single SATA drive vs SAS you may not notice, but with 15 or 20 drives in an ARRAY there will be a very, very noticeable increase in performance with SAS drives.

The deal is expense. How much are you prepared to spend? SAS 15K 300G drives are ~$600 bucks. SATA drives, are less than $300.00 for 750G which is more than double for half the cost. There is reason SAS drives cost more, and that is performance.

I ask the question again.

Do you want performance or do you want disk space?

[bburke]

Makes sense. What is the function of most of your vm's? Are you using 1 array or 2+ for your datastores?

That's great to know about the performance comparison between the PS and the CX500. From what we've heard, the EqualLogic array's don't seem to have a downside.

[doubleH]

i don't think anyone has said that SAS is not faster than SATA. what i have said is that it depends what he is going to put on the arrays. If it's light load stuff he will be happy with SATA. many of us are using EQL SATA arrays and are very happy with their performance/features.

If you found this or any other post helpful please consider the use of the Helpfull/Correct buttons to award points

[Patrick_Miller]

It's not only a performance question, but a reliability question as well. SAS drives are built for a higher MTBF than SATA drives...keep that in mind as well.

If I remember my figures correctly, the "average" IOPS per platter is as follows:

60 for SATA

130 for 10k SAS

180 for 15k SAS

[bburke]

MTBF?

[Patrick_Miller]

Mean Time Between Failure

[seveler1]

I am with another person who asks how many vm's and what is the usage going to be. Here is what we have.

Approx 300 users

Exchange 2003

2 DC's

2 File servers

3-4 SQL servers

3 Citrix

misc others.

We have a total of 31-32 vms with one ps300 and just added a ps5000. The EQL stats show that we are barely touching the SAN. We have three ESX hosts that run about 25% utilization on both cpu and memory. They are dual core amd 2.6 with 16gig of ram. I can say that the EQL is the only way to go. Setup the SAN in 5 minutes. Add another san to the storage group in 5 minutes. Snap shots, replication, thin provisioning, ........ Email me if you want to talk more but, if you are not going to be pushing the SAN very hard then SATA has been good to me. You might want to ask your sales person to let you buy the SATA and if it does not work see if they will swap out with the SAS. My experience with them is they will.

[bburke]

This is great. I haven't been able to demo the EQL utilization stats or reports, but anything is going to be better than the reporting that we get out of the CX300 we have, which is nothing do to the fact that we haven't bought the license to do reporting.

And are you running all of the below servers as vm's? Are you using hardware iSCSI HBAs or software initiators on the hosts? For your exchange and SQL vm's, how are you connecting them to the EQL...the MS software initiators on the vm?

[christianZ]

Have you checked this: http://communities.vmware.com/message/584154

You will see the best throughput with ms iscsi initiator but that costs cpu power.

The best choice for Esx should be the iscsi hba (Qlogic) but don't forget there are problems with support it in HP Servers.

We have here some Linux servers (with Oracle DBs) they're booting from san (with Qlogic hbas) - the performance is good.

There were some issues with these hbas - the newer fw was needed then. There are differences between 3.0.2 (don't use the newest fw with it) and 3.5 - the new Qle 406x are only supported with Esx 3.5.

[happyhammer]

ill agree with Chrstian ive posted a couple of tests on both the PS100e SATA and the PS3900xv SAS 15k. Ignoring the max tests you can see below the difference between the 2 on real life and random, these were both with 4052 HBA's in 3.01 or 3.02

PS100 @Raid50 SATA

RealLife-60%Rand-65%Read......___37_____..........___1302___.........____10____

Random-8k-70%Read.................____42____..........___1183___.........____9____

PS3900XV @Raid10 SAS 15K

RealLife-60%Rand-65%Read...___13.6____.........._____3845__........._____30.04___

Random-8k-70%Read.................__16_____..........__3154____........._____24.6___