ESXi 6.5 issue

thebloyn · 22 Dec 2016 at 11:47

I have used Esxi 5 before, so not new to it, got a new Dell T20 Xeon, 4gb Ram, I have 2 VMs on there using Esxi 6.5, both unbuntu 16.10 server.

One VM is nextcloud, the other runs iRedmail.

However, rather annoyingly, whichever VM was the last one to boot, will hang after half a day or so, the other one is unaffected. I have tried juggling the 4GB between them etc but it keeps happening.

SupraWez · 22 Dec 2016 at 12:11

Have you checked the VM log file located in its directory on the datastore, might give you some idea why its hanging.

VM tools installed?

How many vCPU and how much memory do you have assigned to each VM?

thebloyn · 22 Dec 2016 at 12:46

I have tools installed yes, I will see if I can dig out the log when home, its not working from work.

Each vm has access to all the cores, nextcloud has 2gb and mail 1gb ram leaving the rest to esxi, ive tried 2gb to each and other combos but make no difference

Liquidfox · 22 Dec 2016 at 13:27

When the VM hangs, what's the total memory usage on the host? 1GB was enough for 5.5 with no passthrough however is this still applicable for 6.5? How much memory does the host use with no VM's on?

thebloyn · 22 Dec 2016 at 23:22

Liquidfox · 23 Dec 2016 at 02:11

https://kb.vmware.com/selfservice/m...nguage=en_US&cmd=displayKC&externalId=2000542

SupraWez · 24 Dec 2016 at 15:21

thebloyn said:
Each vm has access to all the cores

This is a common mistake when building virtual environments and an easy mistake to make when moving from the physical environment to virtual, the way ESXi slices up CPU time means this is not optimal, if you only have two VMs you would be far better off assigning half the cores to each VM, even then this would be overkill.

As an example, if you have four real cores and create four VMs each with one vCPU they would all run together unhindered, if you then changed them to having two vCPU each, only two of your VMs can process at anyone time, meaning the other two have to wait for CPU time to become available. Now if you increase all four to have four vCPU only one VM can be operational at anyone time, this is not ideal, this is a simplified example but gives you a rough idea.

Hellsmk2 · 28 Dec 2016 at 12:17

SupraWez is spot on.

If you were to look at the performance metrics for the VMs from the client and select CPU Ready as the only metric, you will probably see it is sky high (CPU Ready tells you how long the VM has been waiting to access the physical CPU's - as a general rule of thumb, anything real-time above 100ms would lead me to look at the VM CPU allocation).