VDI/VMWare View - Best Practice For 600 VM's

[gmgohara]

Hi

We are putting together a 600 user VDI / VMWare view solution which has been sized across 12 x ESX Hosts using NFS (NetApp) for filestore

Proposed design is :

ESX Hosts (running DRS / Vmotion etc)

NetApp Filer Running NFS for VM storage

Each ESX Host would be connected as follows :

1 x Connection to Storage LAN1 (NFS)

1 x Connection to Storage LAN2 (NFS)

1 x Connection to UserLAN1

1 x Connection to UserLAN2

1 x Connection to VMWare Mgr LAN

1 x Connection to HP ILO LAN

The intention is that we have 2 master VM images. Image A is connected to Storage LAN1 / User LAN1 and Image B is connected to Storage LAN 2 / User LAN2

This is to ensure that we get a balance of traffic over the various interfaces. Believe this is best if you want to balance the network traffic when using NFS stores. Is my thinking correct here that you cannot load balance NFS datastores across 2 physical NIC's

What I am also trying to understand is to how you ensure that there is a balance of VM's using each image on each host i.e. Host 1 has 50 VM's (25 of which are Image A / 25 Image B). In essence, we are trying to effectively ensure that we do not get a host which is full of the same image type (meaning that only 1 set of interfaces are used and potentially overloaded). We want to ensure that whenusing DRS that we get an equal split of image types to host (if you see what I mean).

Any advice appreciated

[KFM]

First off the bat:

You cannot use a cluster with more than 8 ESX hosts to support your VDI infrastructure. This will be rectified in a later release of View, but for the current version, it only supports clusters with 8 or less ESX hosts.

Is my thinking correct here that you cannot load balance NFS datastores across 2 physical NIC's

Yes I believe your thinking is correct, unless you have multiple NFS datastores (each with a different IP address) to which you connect to, in which case you can select IP hash and let it choose a preferred vmnic to use. Else if you only have one NFS datastore then the same path will be chosen all the time.

Not sure how you could achieve the image balancing act however. You could always turn off DRS on the cluster and manually place everything until you find a good balance in terms of I/O?

[Adminafterwork]

Hi there,

sorry for not giving you any big help regarding your questions and thinkins,. just wan't to say, that we're highly interessted to see your further steps regarding this project, cause we will go for a similoar project with 3 different vm-images, NFS and around 450-500 users.... thx

[jimjimk]

HI, Load balancing your NFS datastores is a must!! I suppose you could use AntiAffinity rules on your DRS cluster to keep certain machines apart...?

[KFM]

>I suppose you could use AntiAffinity rules on your DRS cluster to keep certain machines apart...?

Not really. If you created an anti-affinity rule to keep apart VMs based on their I/O profile, then you would have ESX hosts with just one type of VM on them which would defeat the purpose of load-balancing VMs of different I/O profiles on your ESX hosts.

[jimjimk]

That's not always a bad Idea. (more often it is) Especially when there is a good administrative reason to separate, remeber this is a VDI infrastructure and not a server infrastructure...