VMware Solutions Discussions

New FAS2040 with Vmware storage design

black_spider
7,440 Views

I just purchased a new FAS2040 with interal SAS drives and a DS4243 with full of 1TB drives in it. It will be primary used by our VMware. Being new to to NetApp system, I am not very familiar with the best practice with VMware. I have a Microsoft Exchange server project that requires to be in the VMware environment. I am trying to design the best way that the heavy I/O won't have much impact on the vms. Here is my delima.

  1. Create an aggregate in the FAS2040 and assign as many disk to it as possible. Create one flexvol and use it as LUN for VMware
  2. Create an aggregate in the FAS2040 and assign as many disk to it as possible. Create multiple smaller flexvol and use them as LUNs for my VMware.
  3. Create muliple smaller aggregate in the FAS2040 and create a flexvol for each aggregate. Make each flexvol as LUN for VMware.

Which one would yeild the best result?

1 ACCEPTED SOLUTION

keitha
7,439 Views

YOu definitly want as many disks as possible in your Aggr so option 3 should be discarded.

What protocol are you using to connect from ESX to the NetApp? If it is NFS a single FlexVol for your VMs would be fine performance wise and maximize your dedupe. If you are using FC/ISCSI you may want smaller LUNs for better performance. It varies but usually 15-20 VMs per LUN max maybe less.  Then you have to decide if you want multiple LUNs in one Volume or one LUN per Volume. One LUN per volume gives you better control for things like replication, Vaulting and SnapShots but Multiple LUNs per Volume will get you better Dedupe. It's a trade off. Both work well.

View solution in original post

10 REPLIES 10

keitha
7,440 Views

YOu definitly want as many disks as possible in your Aggr so option 3 should be discarded.

What protocol are you using to connect from ESX to the NetApp? If it is NFS a single FlexVol for your VMs would be fine performance wise and maximize your dedupe. If you are using FC/ISCSI you may want smaller LUNs for better performance. It varies but usually 15-20 VMs per LUN max maybe less.  Then you have to decide if you want multiple LUNs in one Volume or one LUN per Volume. One LUN per volume gives you better control for things like replication, Vaulting and SnapShots but Multiple LUNs per Volume will get you better Dedupe. It's a trade off. Both work well.

black_spider
7,415 Views

Thanks. I am going to use FC because we purchase a HBA card for each ESX host and fiber switches. I thought FC is much faster than NFS.

johnlockie
7,415 Views

additionally, is this a dual head system?  or single controller? 

i would recommend NFS datastores for VMWare

regarding HA and multipath active/active configurations check this out: http://media.netapp.com/documents/tr-3437.pdf

johnlockie
7,415 Views

The reason I ask is because you may create 2 aggregates in an active/active config, and assign a volume in each aggregate to handle VMWare datastores. This would spread your load to both controllers.

Otherwise, if you have a single controller you should stick to a single aggregate with one volume for VMWare datastore(s).  In fact, I would stick to VMWare guidelines instead of NetApp here (regarding number of VMs per store).  This was already mentioned in the first reply to your post.

I recommend NFS because of A-SIS - dedupe is phenomenal on NFS VMWare datastores, especially for Windows

Now for Exchange you should probably create additional volumes for LUNs to mount logs and database disks to Exchange.  You can use RDMs too.  Depends if you licensed SnapDrive, etc. etc.  Also, did you check out SnapManager for Exchange?

black_spider
7,415 Views

My FAS2040 has dual controller, and it is currently configured as Active-Active, and I am planning to create an aggreate on each controller for better load balancing. As far as using NFS, unfortunately, we already purchased FC for all the ESX hosts. Therefore, we have to use FC.

I was thinking create multiple mailbox servers in vm without using RDM to take advantage of HA. You think using RDM will yeild better performance? We don't have any license on SnapDrive nor SnapManager.

m_lubinski
7,415 Views

Remeber that in active-active configuration, you will divide total number of disks /2 so for "data aggr you will be able to use only 12x 1TB SATA disks. minus 1 disk (at least) as Hotspare, so 11 disks.

You may install Ontap 8.0.1 then you may use iSCSI or FC luns without previous limit of 10-15VM/lun due to VAAI support from Netapp with Ontap8 and vSphere 4.1

I would not use RDM due to its complexity. I heard very good approach for general designs: KISS - Keep It Simple Stupid

Remember to use either Windows 2008 OS or if not possible, make sure your VMDK are aligned (linux+win2003/2000)

p_maniawski
7,415 Views

Have you benchmarked FAS2040? I'm really interrested on it's performance with small (2k) sequential writes. It would really help me with Thread Link : 13413.

m_lubinski
7,415 Views

how do you expect to benchmark fas2040?

with using what tool?

Marek

p_maniawski
7,415 Views

I've been benchmarking "mine" mainly with sqlio (win), but you can also use sqliosim (win), bonnie++ (linux), hdparm (linux), dd (linux) and more. Please take a look at Thread Link : 13413.

m_lubinski
5,618 Views

right now i can't benchmark it, but i have n3400 (fas2040 equivalent) with 48x SATA disks, so I will check it somewhere next week.

Public