Link to home
Start Free TrialLog in
Avatar of Indy_Admin
Indy_AdminFlag for United States of America

asked on

Seeking comments/advise on the following NAS setup.

I am seeking comments and recommendations on the following seutp.  This was in place prior to my enterance to the environment and would like to know what I am facing.

Two IBM x3550 servers (Dual quadcore 3.66Ghz, 32GB, RAID 1 73GB SAS rpm unknown)
HYPER-V running the following
Two Win 2k3 Std / Citrix 4.5 serving 220 users (average 45-55 users concurrently) (All desktop apps are on these servers)
One SQL 2000 server running a customized database (used by approximatly 100 users some light some heavy)

All three virtualized and running on QNAP TS-859U-RP+ Turbo NAS with 4x 600GB SATA 10K drives in RAID 5.

In addition there is a plan to add the following to the Hyper-V setup and NAS:
Exchange 2010 (220 active mailboxes - approx 100 fully utilized)
Upgrading Citirix 4.5 to XenApp 6 utilizing both until upgrade complete.
SQL 2008 R2 Std to replace the above mentioned SQL 2000 install.  Both will operate for approximately 1 year prior to the shutdown of the first.
Accounting Server running a version of SQL express 2005.
2 to 3 test servers running misc apps.

My concerns are:
# of Disks being used in NAS (Disk Queuing)
Virtualizing SQL to the NAS
Virtualizing Exchange to the NAS
When running batch reporting our entire system grinds to a halt until the batch job is done.

Please advise.  Greatly appreciated.

Avatar of Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Flag of United Kingdom of Great Britain and Northern Ireland image

RAID 10 would be better.

Are you attaching via iSCSI?
Avatar of Indy_Admin

ASKER

The NAS has iSCSI targeting service available and I beleive we are using that.

Is it a major concern is that I have that much read and write data ALL hitting on one logical drive regardless of the array type?  Either way the sytem was setup as a RAID 5.  While I could mirror the 4 drive RAID 5 I don't think that would help with my read write which is where I think the issue/bottleneck will continue to be a problem.

FYI the info on the QNAP devices can be found at http://www.qnap.com/USEng/pro_detail_feature.asp?p_id=185
RAID 10 would provide better read and write performance, but you are limited in what you are going to get out of four 7,200 rpm SATA disks.
Enabling Jumbo Frames on your network interface, Physical Switches and NAS may also help with throughput.
So do you think that I can run a SQL 2008 VM, a Exchange 2010 VM, two Citrix VMs running all apps, a SQL Express VM, and test servers all from this NAS without serious performance issues even with Jumbo Frames enabled?  I would think that the drive IO would choke the system let alone the network traffic.

The drives will be supporting the OSes of all machines, the paging files of all machines, the log files from the SQL and Exchange machines, and the read and write database transactions for the SQL and Exchange machines.
I think it's time to start looking at a larger NAS or SAN.
I have typically been advised to not run SQL or Exchange in a SAN or NAS environment unless they were fiber connected.  Can you comment on that?
Fiber or iSCSI it does not matter, it depends on how many IOPS you NAS or SAN can provide.

You could be running iSCSI over 10GB ethernet whicn is potentially faster than Fibre.

There are many iSCSI based solutions using Exchange and SQL.

Disks in the SAN/NAS are the key.
I would assume then that you also beleive that having a maximum of 8 disks is not enough to sucessfully run the above configuration?  Especially with 4 drives already being configured as a RAID 5?
if you were using 8 - SAS, 10,000rpm disks.

SATA 7,200rpm disks are the devil, you will always have performance issues.

more disk = more spindles = more performance.

here are some example figures from SATA 7,200rpm disks

2 disks - 84.5 MB/s Write,  150 MB/s Read
4 disks - 196 MB/s Write, 276 MB/s Read
8 disks 212 MB/s Write,  287 MB/s Read

more spindles, more performance
agreed.  Also, Seperation of jobs from spindles is important. (OS, logs, paiging, Database read & write).  Would you agree?
it's an old fashioned thoery, we think, as long as you have a very fast datastore, that's presented to OS, it should not make any difference.

That was the thinking with physical machines.

If you have a very large SAN of many many spindles, your storage LUNs are carved out of a massive volume, striped across many storage enclosures of 14 disks.
Love being considered old fashion.  lol

What you are saying makes alot of sense if indeed you have 14 disks to spread the data over, however even with 14 disks removing the paging and writing (logs) from the reading (DB) would still enhance performance wouldn't it?
Yes...... IN THEORY

IF you had another 14 disk SAN to put those on.

But if you are reducing your 14 disk just so you can create a seperate array to present another lun to create another volume to store you logs on then no, because you've just degraded your original 14 disks......

But where would you put the paging files and writing (logs)?

if you had a very large NetApp Filer with

A 3U Chassis - NetApp DS14MK2, 14 shelves, in a 42U rack, driven by a NetApp filer head, High performance RAID 6 or RAID 4.

196 spindles and disks. The LUNs which area presented to the servers are carved out of a pool or storage that is stripped across 196 spindles.

Removing the paging files and logs, may increase performance, but where would you put them?

On what!

You could split the shelves, but less spindles less performance, so performance you gain, is performance you would lose!?

SOLUTION
Avatar of geekgonewild
geekgonewild

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
Geekgonewild,

Does Hyper-V support NFS Mounting?

It is my understanding from the posts I have read that it does not.  In addition I already have a RAID 5 in place which leaves only 4 drives left in the array.  I have found that VMWare, which is the world I came from, is a completely different ball game than Hyper-V when it comes to resourse allocation and availability.  Thank you, however for your post.

Some but not all of the post follow:
http://www.windowsitpro.com/article/nfs/q-does-hyper-v-support-network-file-system-nfs-
http://thestoragearchitect.com/2011/06/02/why-does-microsoft-hyper-v-not-support-nfs/
http://social.technet.microsoft.com/Forums/en-US/virtualmachinemanager/thread/789f8b5b-fba5-4a8e-9cd9-6329d36830ca
http://social.technet.microsoft.com/Forums/en-US/winserverhyperv/thread/2160e204-0ea8-40d3-bf9e-fed6c87c0fc5
http://social.technet.microsoft.com/Forums/en-US/winserverhyperv/thread/5868acd4-3c1e-4883-a799-b25ba84a0d6e
hanccocka,

A bit over the top dont you think???

I agree that taking away spindles would reduce performance, however, you cannot read and write to the same spindle at the same time so moving the logs away at the cost of a couple of disks another array would avoid that unnecessary collission and reduce disk queuing.  In a 14 drive array, how much do you think you would loose by pulling a couple of drives out for logging?

The answer probably resides in how much writing vs reading the system is doing simultaneously.

Either way, my original post was centered around a currently populated 4 disk array not being able to handle batch reporting when it only has two citirix presetnation servers serving all desktop apps and a transactional DB server.  I dont think that even if I were able to change the system to a RAID 6 (7+1), adding the additional SQL box, Exchange Box, Accounting Box, and test servers would be feasible...
ASKER CERTIFIED SOLUTION
Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
The point, I was trying to make is the DBAs come to the Storage Engineers, with tail between legs asking:-

Can I have a VM of the following:-

RAID 1 - OS (two disks)
RAID 1 - DB (database)
RAID 1 - DB (logs)
RAID 5 - Page file

and it was demonstrated that what they asked for was slower, than 5 LUNs carved out of the existing Storage Array.
Point well taken.  Thank you for your help.