1

I have servers running a few different iSCSI SAN units. I was trying to balance IO requests to the datastores, so I started changing the "Limit - IOPS" on my servers from unlimited to 100-400 depending on what the servers role is.

In doing this, it causes that specific VM to run VERY slowing. For exsample "unlimited" IOPS, IE will open in 1 second. Iops limited to 150, IE will take 10-12 or more to open on that server. I have verified this to be a issue on two completely different iSCSI SAN units as well.

Any ideas what I am doing wrong.

Litzner
  • 405
  • 1
  • 5
  • 17

2 Answers2

6

I think this is doing exactly what you're asking it to. You're throttling the I/O capabilities and profile of your VMs, causing them to perform poorly.

Don't do this.

Micromanaging resources at the individual VM level can have unintended consequences and will complicate future troubleshooting.

My recommendation is to NOT try to outsmart the VMware scheduling and resource prioritization algorithms and allow things to run unlimited until you really do have a storage contention issue.

Was there a problem that prompted you to try to balance the SAN usage in the first place?

ewwhite
  • 197,159
  • 92
  • 443
  • 809
  • I am was having latency issues with my SAN units from time to time, where latency would spike from 8ms to 300ms. I was thinking this was cause by some VMs eating up more then their fair share of IOPs. So I was experimenting to remedy the problem. – Litzner Aug 14 '14 at 18:07
  • @Litzner That would probably be a SAN-level issue. The better question is whether those latency spikes have a negative impact on your users/applications. – ewwhite Aug 14 '14 at 18:16
  • IO throttling causes latency spikes. Because an IO gets 'stalled' when it hits a throttle. Most applications get upset by this, and run really badly. Sometimes you need to do it, to prevent link saturation and knock on impact - throttling upstream so you don't cause contention on your core switches, for example. But fundamentally - you're not fixing anything, you're just choosing whether you'd rather cut off an arm or a leg. – Sobrique Aug 19 '14 at 13:29
0

I don't believe this question has been answered, I recently found another article that discusses this issue and a work around for 5.5 users. http://blog.igics.com/2014/07/heads-up-vmware-sioc-virtual-disk-iops.html

Bob
  • 1
  • 1
    Whilst this may theoretically answer the question, please [provide context around links](http://serverfault.com/help/how-to-answer) so others will have some idea what it is and why it’s there. If possible summarise or quote the most relevant part of an important link, in case the target site is unreachable or goes permanently offline. – HBruijn Apr 09 '15 at 16:33
  • It was a bad question to start with. – ewwhite Apr 09 '15 at 16:39