I have a request from my client to enable Virtual Machine condition counter (in alarms) >>> VM Max Total Disk Latency (ms).
VMware states by default the values are 50 milliseconds as warning and 75 milliseconds as error. Snapshot attached
I googling on what to keep the value for this counter and then I saw link VMware vSphere 4 - ESX and vCenter Server
It states Low value for the counter is – 0-2 seconds ............ Moderate value is – 2-6 seconds .......................... High value is – More than 6 seconds >>>> Though these values were for ESX 4. Snapshot attached
1 Second = 1000 Milliseconds
As per the ESX4 link till 2000 milliseconds if the latency goes for VM disk this should considered as Low. I was unable to find the documentation for vSphere 5.
Please let me know is the ESX4 documentation should have been modified or the settings that come up as default for 50 are fine? What values you guys have applied in your environments and what value should I apply here
Thanks in advance
I would say if your latency stays about 20-30 seconds consistently you probably are going to start seeing problems. I would set alarms at 30 for a period of time, but it really depends on the environment.
Thanks for replying Mark ...
Just wanted to check the link which points to ESX4 documentation suggests that 0 to 2 seconds is fine for latency that means 2000 milliseconds. That needs updation ?
Sorry I meant to say 20-30 ms isn't good not 20 to 30 seconds. I am not sure on the documentation, but I've been doing VMware work and design for 8 years so I know from experience. My guess is that is a typo and they meant to say MS. I mean ideally you would like to stay under 10 ms at all times and as low as possible
This is a good document:
Thanks Mark for sharing the links. I read all the three parts and it is very good. I have couple of questions that i wanted to ask
The document said if the latency is constantly high for davg and is above 30 ms then we need to check. What do we mean by constantly high do we have a number in terms of seconds/mins what this will be ideally
Second .. in my enviornment if vm disk latency ( which is davg + kavg) for a scsi vmdk is above 100 or 150 ms for 10 mins and then comes back to normal. Is this something that we shoukd be worried about or can this be ignored.
Thanks in advance
A lot of that depends on the type of workload the servers have. 100-150 for 10 minutes seems high to me and I would try ot determine what is causing that, but if it only happens rarely probably acceptable. If it is during backups or something of that nature that could be the cause. Honestly, there are too many variables to give a straight answer. If the users aren't experiencing any slowness than it may not be an issue.