VDI / Terminal services

[Dave_O]

Update on progress of VDI

 

OK so we were going to buy Atlantis ILIO (diskless) for 50 desktops, which turned out to be about £3,000. Expensive but seemed worth it. After much testing we found that we needed the persistent disk version, asked for a re-quote expecting it to be perhaps a bit more something like £3,200. Nope £7,000!!!! So after many expletives we told Atlantis where to go. I then refocused on Nexenta. Nice technology, not quite as effective as Atlantis but seems to work OK. What was even better is that there is a community version (not the VDI version - which lets face it is just the storage with some bells and whistles which I don't need) which is free. So I put that on and here we are. Will report back on performance later after half term. I do like NexentaStor, simple and effective. Lets see what happens now.

 

Also looking at the Microsoft licensing of VDI. Bit like Pandora's box. More on that when Paid_Peanuts gets the final word from MS.

[Dave_O]

Update II (this time it's personal)

 

Ended up ditching NexentaStor community as we got better performance from raw disks. The final configuration we ended up with was:-

 

Dell R720, 256Gb memory, 2 x 2GHz 8 core processors, ESXi boot from SD cards, 12 x 146Gb 15K SAS in a R10 (usable capacity 680Gb), 4 x 300Gb SSDs arranged in 2 x R1 300Gb

 

VMs are Windows 7 2Gb memory, 2 vCPUs optimized for VDI

 

All pools are thin provisioned (VMware View 5.1)

 

VM pools are

 

- students, 45 VMs - gold image and replicas on one of the SSD based drives, VMs on SAS drives

 

- staff, 25VMs - gold image and replicas on one of the SSD based drives, VMs on other SSD based drive

 

With 15 students logged in student pool IO Meter (512B 25% read) gives 4720 iops

With no staff logged in staff pool IO Meter (512B 25% read) gives 4923 iops

 

With these 70 machines running on the 1 ESX server it uses just over 140Gb of memory (as expected) and approximately 9000 MHz CPU (steady state). With an approximate density of 4.4 VMs per core. The SAS drives are capable of delivering just over 2000 iops in their present arrangement giving a worst case of 28 iops per VM but as you can see above this is unlikely to happen in normal usage. I found that from an off state to steady state powering on all VMs as quickly as possible take approximately 30 minutes, with login possible before that but performance is slow. Normal login from steady state for a class of approximately 25 students (using thin clients to access the VMs) is 65 seconds (roaming profile, no profile on the box) - compared to 45 seconds for equivalent physical PC in same state. How this would scale for more than 25 simultaneous logins I don't know but as long as class logins were staggered by 2 minutes there should be no problems.

 

There seems to be very little difference in the reported usability of either machine pool. I remain unconvinced about the value of using SSDs as oppose to raided SAS drives in a school context.

 

I believe this configuration will support up to 100 VMs with little in the way of performance degradation. Cost of all the server side hardware is around £6000 ie £60 per VM

Edited June 3, 2014 by Dave_O

[PotNoodleTech]

Interesting!

 

Your IOmeter results are from within the VM?

[Dave_O]

Your IOmeter results are from within the VM? - YES