NETAPP Aggregate Throughput

[Sparrohawk]

Looking for some insight surrounding Netapp (3140 Filer), ESX3.5 storage via NFS.

How best to configure the "datastore" on the NFS volume to get the best aggregate throughput.

Does ESX create a one to one connection to the NFS datastore? Many to Many? If I assign multiple paths in NFS to the same ESX host can I improve throughput? OR create smaller NFS volumes and present them to ESX that way. Would this create multiple i/o queues? ( of course I am still restricted to the bandwidth relevant to the network backbone and NIC )

Example: 1 TB NFS volume presented to ESX host using for different share points to the same NFS volume.

or

5 x 200 GB NFS volumes presented to the same ESX host.

[mail_vijayar]

Hi,

NetApp says 500GB is a sweet spot to get the best NFS performance from a LUN mounted on a ESX, if you currently have a 1 TB aggr i would recommend splitting it into 2 volumes of 500GB, we can create multiple NFS stores from each volume; NFS on a NetApp is know to have the best performance than on any other storage and is easier to manage and gives more flexibility than block level file system.

Regards,

Vijay

[vmroyale]

Hello.

Make sure you read the NetApp and VMware Virtual Infrastructure 3 Storage Best Practices document. It will answer all of your questions, and probably lead you to more as well!

Good Luck!

[Sparrohawk]

Thanks guys, but I will read the document. I am looking for some gritty details regarding how the queue connections work based upon one single NFS volume or multiple ones...not just what is best, but how and why.

[vmroyale]

It might be worth trying to get in touch with your local NetApp SE to see what he or she has to say. The forums over at NetApp NOW might be worth investigating as well.

Whatever you find, will you please post your findings back here?

Thanks.

[admin]

Use RAID-DP with as many spindles as possible under your aggregates. Build your NFS volumes from there and dont thin provision them. The connections by default will be one to one but with some careful planning of your networks you can acheive redundancy with load balancing from each ESX host. As vmroyale states VI3 on NetApp Best Practices is a must read. It is TR 3428 and you need to review your networking options based on your physical switch infrastructure and whether your switches support cross stack etherchannel (Cisco 3750). Because NFS is a NAS protocol it does not have the same shared filesystem queue as VMFS and actually each virtual machine disk file has its own queue. Your datastores get created on each host in the cluster. Because there is no metadata updates and subsequent scsi2 reservations on the entire LUN you will probably notice you are able to scale VMs per datastores a little bigger assuming you have the throughput on the network configured correctly. This in turn leads to the ability to make your datastores a little bigger which is less administrative overhead. Assuming you have a NetApp cluster, split your datastores (if possible) across controllers to increase available throughput. I have customers running 1.5TB datastores for Virtual Machine Storage on NFS volumes against a NetApp 3050 cluster.