Network Latency after Recomposing

[HLJTechSupport]

Hello all.

We are currently running View 5.2.0 and will be upgrading to 5.3.2 this weekend.   We have 250 32-bit Windows 7 VM's on our network divided up into 8 pools all being linked persistent clones using Persona Management.  This being said, we have a huge network latency issue when users login the next day after a company-wide recompose of all vm's.  The latency starts when everyone logs into their vm's at the same time around 9:00am and the issue finally subsides about 1 1/2 - 2 hours later.  I'm pretty sure this is due to the fact that everyone is creating a new local profile on every VM at the same exact time putting a lot of stress on the DC's to get profile information.  How does everyone get around this issue? The only way I have been able to get around this is to only recompose groups of 30-40 VM's in one night and this is fine until we come across a major upgrade like the one I will be performing this weekend.

A little side question:  Is it "OK" to upgrade to View 5.3.2 and have a View Agent of 5.3.0? Only ask because I would like to slowly recompose all machines unless I can get the issue above fixed in time.

[vcpguy]

Storage is one of the major reason for performance issue. Could you please tell us what kind of storage you are using? RAID configuration? Using Persona Management should take care of User login issue. Your entire user profile is not downloaded during the login. It uses just-in-time file retrieval of user data reducing I/O operations during login and logout operations.

It could be some other issue. How is your anti-virus configured? Do you have vCOPS or Liquidware UX for performance Management?

[HLJTechSupport]

Thank you for the reply.  We are using an EMC 15K SAS Raid 10. Is it possible that I do not have Persona set up correctly?  I attached my Persona settings if you could view them and just double check I have it set up correctly. As for Antivirus we are currently using Webroot which is set up on each VM.  We are going to be implementing Kasperky with VMware vShield in the near future.  We do not use vCOPS or Liquidware UX.

Thank You!

Ben

[vcpguy]

How many VMs are configured per LUN ? Why don't you download trail version of Liquiware UX and check the issue. You can use trail version.

I would have suggested vCOPS but it is little bit more complicated.

[HLJTechSupport]

I think you hit the nail on the head....  We only have 6 Datastores (one LUN per datastore) set up with 253 VM's.  So if I understand this correct, VMware recommends about 10 vm's per LUN (depending on the environment of course) so I would have to create about 25 Datastores to accomplish this?

I have never heard of Liquiware UX but I did a google search and it definitely looks very helpful, I will have to try that out.

Excuse my ignorance, I had to set up Horizon with little to no knowledge in about 6 months to pull off a Windows 7 upgrade before XP expired so I obviously had a lot of trial an error and don't know the in's and out's of everything.

[vcpguy]

I am glad to help you. I think you can use some advance tool like esxtop and see when the users login how is the VM/host/disk performance. This will give you a good idea if we have a problem with the disk.

I don't remember reading 10 VMs per datastore. I think for Flash based storage it is around 160 VMs.

[HLJTechSupport · Accepted solution]

We found the answer to our problem.  We did esxtop logging with 2 VMA's and noticed our IOPS is way to high for what we can handle via our x2 EMC VNXE 3150.  We are currently at 8,000+ IOPS during a normal workload and our VNXE can only handle about 5,000.  The vendor who recommended the EMC device is now buying back our existing x2 VNXE 3150's and we are buying a VNX 5200 loaded with flash which we are hoping to fix all performance issues.