Slow VM's

ashrobbo

ashrobbo

ashrobbo

Associate
Joined
3 May 2009
Posts
805
Hi All,

Trying to troubleshoot a slow citrix server for a customer which runs on an esxi 5.1 box.

Its an IBM 3300, Raid 5 (3 disks), two 6 core xeon cpu's (12 with HT, 24 entire machine) 32GB ram.

RUnning 3 servers on this box, Citrix has 16GB of RAM but constantly has high CPU usage.

It did have 2 vCPU with 2 cores (same as the other 2 vm's).

I have reduced this down to 1 vcpu with 4 cores and the CPU usage seems lower, however its still slow. The disk queue on the server is high at times (sometimes over 4/5) so im guessing its the raid disks not keeping up.

I did read that i may have over provisioned the CPU, so im wondering whether to reduce all to 1 processor and 2 cores? Shocked that 2 x 6 core processors are having issues so must be my config.

Ive turned off any guest power management using the vsphere client.

Any tips for troubleshooting this? The other server is a DC/File and another one runs web filtering and AV amongst a few smaller bespoke apps. If I can improve performance on all it would be good however citrix is the priority.

CPU
PU37Fdy.jpg


DISK
nPp5YdG.jpg


Thanks,

Ash
 
2xvcpu + 2 cores and 1xvcpu + 4 cores is the same thing in performance terms :p

If the cpu is high in the citrix guest then troubleshoot it in the same way you would a physical server. Check the processes that take up the most cpu time. Or perhaps there are simply too many users?
 
The cpu is now much better after i reduced number of vcpu's and increased core to 4 on that 1 vcpu, processes are ok and it is the occasional spike when user logs on.

however im unsure whether to tell of itc disk or cpu related bottlenecks.
 
Is it just a single esxi host or is it hooked into vcenter?

Either way look at the performance data tab in the vsphere client for that host.
 
3 disks in RAID5... what disks?

RAID5 is usually a Very Bad Thing for performance in almost any case, let alone so few spindles in a VMWare environment...

I/O is probably the most critical factor for VM performance after RAM.
 
3 raid 5 disks, sata 7.2k - not my choice.

Nothing stands out from those screenshots? no bottlenecks visible? I always thought it was disks I just wanted to figure out how bad it really is but vmware links are contradicting!
 
To determine how many cpu sockets you need tp assign to a vm you need to look at how much ram you're using. If you're using more ram than what's allocated to one physical cpu on the mobo you'll need a second socket on the vm. This may help explain things better.

Allocating more vcpus to a vm than is needed will reduce the performance, the more vcpus you have the bigger the space the physical cpu will need to find to accommodate the cpu cycle. For example 1 vcpu needs 1 space 6 vcpus need 6

3x 7200 rpm disks = ~ 225 IOPs before the RAID penalty

MW
 
you say it has constant high cpu usage? well thats probably your issue whats using the CPU up?

remember servers are slow pieces of junk if you had cpu hungery apps running on i5's 3.4 workstations then stick 10 of these people onto citrix it aint gonna be happy...

12 core is not a lot...

you are missing all the vital info - how many users what apps have you checked the CPU's are jsut not maxed out?
 
Sorry lots of questions.

Didn't have much time today. To answer a few questions

- 8 users.
- it did have high CPU, CPU is now much lower on the host. The odd spike when people launch sage chrome etc.
- I suggested 15k sas disks but wanted to be sure disk was the bottleneck. I'm seeing disks queue length spike on guest too.
- thanks for links. Ill take a look.
- yes hardware raid with cache (but no cache battery!!!)

Would adding a 4th disk improve by much?
 
Any idea how much this is being hit? tbh its most from all the information given its most likely to be the storage is the bottle neck. Are these enterprise sata or cornershop sata? whats the max capacity for physical disks in the unit? sorry I dont know ibm hw these days
 
ashrobbo said:
Sorry lots of questions.

Didn't have much time today. To answer a few questions

- 8 users.
- it did have high CPU, CPU is now much lower on the host. The odd spike when people launch sage chrome etc.
- I suggested 15k sas disks but wanted to be sure disk was the bottleneck. I'm seeing disks queue length spike on guest too.
- thanks for links. Ill take a look.
- yes hardware raid with cache (but no cache battery!!!)

Would adding a 4th disk improve by much?
Click to expand...

A fourth disk would increase the IOPs, especially if you have 4x 15k disks, that would increase your IOPs to nearer 700 (before RAID penalty).

Before commiting to the upgrade, your best bet is to find out what the storage requirements are for Citirx running on vmware 5.1

The battery will save and store data that's written to the cache but not yet written to disk in the event of a server failure, so it can be recovered.

MW
 
I know the battery is the case with dell/HP however our reseller is adamant this isn't the case with the IBM raid card M5015.

Can't find any tech docs to confirm this though.

Anybody know?
 
Apologies its an M5110, but again adament it supports write back without a battery, im wondering whether I need to 'force' it to do this though. Trying to get LSI storage manager to see the card but because its an esxi 5.1 host it wont see it :(
 
You must log in or register to reply here.
Back
Top Bottom