Planning migration from DAS to iscsi SAN

[gbras]

Hi,

we have an ESX 3.5 physical server running 6 VM (1 win 2003 SBS and 5 linux) from local SAS disks

It's a dell 2xquad core 2.66GHz

now we're adding a second physical server (dell 2xquad core 2.5Ghz) and an ISCSI SAN: we would like to install ESXi on the second server and move all the vm storage on iscsi, so if the first server goes down, in minutes we can manually configure the second one to run all the vms (we are a SMB and at the moment cannot afford enterprise/vmotion license).

I'm asking for advice because we would virtualize an old physical win2003 terminal server as well (with 10-15 users, working mainly with MS office applications) using the second new physical server.

it's the first time we install a SAN (it's a dell MD300i, with 10 146GB SAS disks, plus 5 1TB SATA disks for snapshots and backup) and we want to get the best performance out of it. Reading other posts I understand that disks are very important in virtualizing a terminal server and having good performance out of it.

So, with that type of SAN and with our workload, should we define a unique RAID 5disk group with 9 disks and 1 hot spare, with a big virtual disk on it to create one or more VMFS datastores?

or is it better to create 2 disks groups:

-1 with two RAID1 disks for terminal server VM

-1 with seven RAID5 disks for all other VM

with a global hot spare disk?

other general suggestions for the migration we're going to do?

Thanks in advance

Guido

[kjb007]

I would run some perfmon tests to see how much disk I/O your terminal server is doing right now, to see if you need to take away storage from your pool. You will lose ~ 100 GB overall using the mirror, but you will get better performance. What is "better" depends on your needs, and good performance is a relative measure. And since you have a new server, you can create a terminal server vm on your new storage and run IOmeter /sqlio to see what kind of I/O you can get out of your RAID5 vs a RAID1 diskset, and see if it meets/exceeds your current configuration and requirements.

Since you can't migrate without virtual center, it may be worth the time to "evaluate" virtual center features such as migration during your move. It would also give you a means to create templates/clones which may be helpful during your migration. Otherwise, you'll be downloading/uploading files using scp/datastore browser, and that can be cumbersome.

-KjB

[khughes]

You think a RAID1 would be faster than a RAID5? I kind of find that hard to believe, but I have been wrong before.

• Kyle

[kjb007]

Depends on your channel, writing a parity can be slower than writing to volume which does internal controller channel-based mirroring. Reads are almost certainly faster on a mirror as well. But, again, faster/slower are relative terms, and the difference will show up only when you are trying to push a large amount of I/O. Personally, I would use a RAID5 and create a bigger pool.

-KjB

[khughes]

I would think the RAID5 would be better as well. 10-15 TS connections isn't that high. On our SAN we have a couple PS servers with around 20-30 users per LUN using RAID5 and no I/O issues. I can't seem to think a 10-15 user TS server hosting office apps would cause that big of a load.

• Kyle

[Texiwill]

Hello,

RAID1 is faster than RAID5 if you have raid systems of equal # of spindles. Which is either 4 or 6 disk LUNs as over 7 disks in RAID5 has performance issues in some cases (depends on the array mainly). However, note that in general performance of arrays depends more on spindles than anything else.

.

Best regards,

Edward L. Haletky

VMware Communities User Moderator

====

Author of the book 'VMWare ESX Server in the Enterprise: Planning and Securing Virtualization Servers', Copyright 2008 Pearson Education.

Blue Gears and SearchVMware Pro Blogs: http://www.astroarch.com/wiki/index.php/Blog_Roll

Top Virtualization Security Links: http://www.astroarch.com/wiki/index.php/Top_Virtualization_Security_Links

[gbras]

Thanks to all of you for the replying my question.

Actually I'm not focusing on the raid level to choose, but on determining if I should reserve a couple of disks for exclusive use of the VM running terminal server (2 disks, raid 1)

Guido

[kjb007]

Your I/O requirements do not designate the terminal server as a reason to split your datastore. If your business requirements designate it as such, then that is a different story. Is this vm more important than your others that you need to segregate it into its own disk protection pool? If not, I'd add it to a common pool.

-KjB

[gbras]

With only one big iSCSI datastore shared between these VMs running on two physical servers:

-1 win2003 small business server with exchange and SQL server

-1 win2003 Terminal server

-5/8 linux servers (web, ftp, 2nd mail server, file library... etc)

and 15 users, aren't we going to have too much contention on the VMFS because of locking and scsi reservation issues?

Thank you

Guido

[muczyn2003]

Hi,

Very interesting topic !

We just bought the same unit MD3000i with 14 x 300GB SAS 15 K disk, we plan to use this unit for DR site for approximately 100 users.

Our Dr Site consist of 3 x esx standard, VC and shared storage (MD 3000i), it will host

Production site disk configuration:

2 x citrix servers PS 4.0 raid 1

1 x Domain Controller raid 5

1 x Exchange 2007 raid 5

1 x SQL 2005 raid 5 and raid 1

1 x file server raid 5

1 x apps server raid 1

1 x web server raid 1

1 x citrix web interface raid 1

DR site disk configuration

14 x disks in RAID 10 + one hot spare disk

We were advised to create one big storage array in RAID 10 and put all the virtual disks on that. I am not sure if this is the best disk configuration? I was thinking about grouping the exchange server and sql server and putting them on raid 5, then citrix servers on raid 1, and the rest of the servers on another raid 5 so we have:

Raid 5 6 disks (exchange 2007 and sql 2005)

Raid 1 2 disks ( 2 citrix servers)

Raid 5 6 disks ( web servers , file server and app server)

What do you think?

[kjb007]

I think we've been trying to answer a couple of different questions in this thread. One was RAID1 vs RAID5. And, I think we've answered that. I'd create a single RAID5 set. I would also split your logical drives in that RAID set.

The next question would be how to split your LUNs for best vm performance. That does not follow the same logic. The typical answer on LUN size for vm datastores is 300-500 GB or roughtly 15-20 vm's per lun. Since you're still under that, in terms of number of virtual machines, I don't think you'll run into that issue.

Now, how large are your vm's themselves? How much storage is required per vm? As a general rule, if any of my vm's require more than 300-400 GB for storage, I do not take that space out of my datastore, and instead use an RDM for that bit of storage instead.

Lastly, even if you had all of it coming out of the same datastore, in your scenario, I don't think you would run into a locking/reservation issue, unless you greatly increased the number of vm's and/or the number of physical hosts attaching to those vm's.

-KjB

[muczyn2003]

thanks for the advise,

so if i create one big RAID 5 from 14 x 300Gb and then create 4 LUNs and assign each LUN as a data store in vmware will this setup handle 100 users load?we dont want to experience any lock issue or perfomance degradation ,

total space RAID 5 14x 300= 3.7 TB

LUN#1 Exchange 2007 300GB 1 server only

LUN#2 SQL 2005 and SQL 2000 500GB 2 servers

LUN#3 2 x Citrix Servers, 1 x Citrix WI, 1 x Web interface, 1 x mail marshal and web marshal server, 1 x CHONE 400GB 5 servers

LUN#4 1 x Domain Controller and file server 500GB 1 -2 servers

After creating LUNs around 2 tb shoudl left on MD3000i

[kjb007]

Wow, we have more paths along this thread than I noticed. My previous post was for gbras. For you, since you have 14 disks, I would not create one big RAID set out of that. I'd split your disks into two RAID5 sets. Hopefully, you have a spare as well, else, I would mark one disk as a spare as well. Then, split those into the LUNs you have already defined.

-KjB

[gbras]

Let me check if I understand correctly:

-it's best to create a single large RAID5 physical volume, with 9 disks and 1 hot spare.

-then "partition" that large physical volume in two logical volumes ("I would also split your logical drives in that RAID set."): one logical volume for terminal server VM (approx 100GB) and one for all the other VMs (approx total 1TB).

-Create a LUN (->VMFS datastore) on each of the two logical volumes

Is this ok?

Thank you

Guido

PS

I actually have a VM with large disks (400GB + 50 + 30 + 20 + 10, it's our main server, SBS 2003): I'd prefer not to use RDM because vdmks are easier to backup and recover (I can snapshot them and make a local copy even on a portable usb HD; and If I need to recover a single file I can create a VM on a temp phisical server, attach partable usb hard disk, create a VM using that disk and recover files)

[kjb007 · Accepted solution]

Yes, that is what I would do.

-KjB