VM limitations

[rvtech]

At the moment, the limitation is based on CPU/Core, RAM and harddrive usage, correct?

Are there any other limiters such as IO, user beans, network etc?

Is overselling of RAM/CPU is possible, and what the consequences are of using more RAM than allotted?

Will the VM slow down, are processes killed (something that openvz does afaik)?

 

[martin]

please use the search function of the forum, most questions could be answered this way.

if there is still something unclear, post again. but not multiple questions in one thread to keep it clear and easy to understand for others.

 

[rvtech]

The answers I found are still not very clear:

Yes, as long as the VM does not use more that 8GB. But overcommiting resources is a bad practice IMO and can lead to unexpected errors.

What kind of unexpected errors? OpenVZ kills processes to free ram, incase of 100% ram usage. What does KVM?

Hoping for a clearer answer..

 

[martin]

The answers I found are still not very clear:



What kind of unexpected errors? OpenVZ kills processes to free ram, incase of 100% ram usage. What does KVM?

Hoping for a clearer answer..

this is not only the case on virtualized servers. if you run out of memory (and swap) the system stops processes. this is the case on real hardware, on openvz and also on KVM VM´s and on all other virtualization technologies.

(depending on the guest OS you got unexpected results. Linux guest stops processes, windows guest just crashes most times.)

 

[jcpham]

Overcommiting RAM is bad

It is possible to over commit RAM, just learned this by doing so accidentally. PVE host on poweredge 2900, 2 KVM guests, both windows - SBS 2008 and Server std. 2008.

Server has 8gb of ram total, but a little less actually available to OS (PVE). The VMs were not setup at the same time and by different people, so we had actually committed 8gb of RAM total between just the 2 VMs.

A nightly vzdump to external media would cause the host and both VMs to go completely unresponsive and require a hard reboot.

 

[tom]

It is possible to over commit RAM, just learned this by doing so accidentally. PVE host on poweredge 2900, 2 KVM guests, both windows - SBS 2008 and Server std. 2008.

Server has 8gb of ram total, but a little less actually available to OS (PVE). The VMs were not setup at the same time and by different people, so we had actually committed 8gb of RAM total between just the 2 VMs.

A nightly vzdump to external media would cause the host and both VMs to go completely unresponsive and require a hard reboot.

you can configure VM´s with as many RAM as you like but if you do not have this on the machine and you start to many VM´s the system will perform badly. currently we do not check the available RAM, but it´s already on the todo list (means if you power on and the RAM is already used you got a warning).

your vzdump issue:
pls provide more details, vzdump email logs, etc.

• which mode do you use?
• where is the backup target (local or remote), enough space?
• how big are you VMs?
• etc.

 

[jcpham]

vzdump reply

thanks for the reply tom. I wasn't expecting it. First off, I would like to say that I wasn't "downing" proxmox in any way, I love it's packaging and KVM/VZ seem to blow the other hypervisors out of the water right now. We use it on our production equipment and recommend it to clients.

A little background on what our specific problem was and how it was fixed. I'm not going to be able to provide logs, we've fixed the problem already and it's been awhile. I said before we had 2 VM's and we think we overcommited the RAM. Server itself has 8 cores, the VMs were only using 4 total. 2 cores for each VM.

vzdump snapshot would run nightly with compression and backup to local USB drive. the 2 VMs totaled about 170gb uncompressed. The external drive is 1tb and the client rotates the drives offsite every day. Precisely when vzdump would start, both VMs would stop responding. As in, you can't log into them and the no events would be logged on both VMs until they were stopped and restarted. Both VMs would also blue screen, the blue screen error would say something about an unresponsive processor. There were 2 instances where the host went unresponsive also. I could ping the host, but no ssh or web management. rebooting the host would bring everything back to life.

Ultimately, we turned off compression on the vzdump backup and I lowered the total available RAM to both VMs and we haven't had the problem since.

 

[dietmar]

Ultimately, we turned off compression on the vzdump backup and I lowered the total available RAM to both VMs and we haven't had the problem since.

Most likely, you have a slow disk system? When you overcommit RAM, server starts to swap, resulting in increased IO activities. When you then start vzdump in snapshot mode, IO gets a real bottlneck.

-Dietmar

 

[impire]

you can configure VM´s with as many RAM as you like but if you do not have this on the machine and you start to many VM´s the system will perform badly. currently we do not check the available RAM, but it´s already on the todo list (means if you power on and the RAM is already used you got a warning).

your vzdump issue:
pls provide more details, vzdump email logs, etc.

• which mode do you use?
• where is the backup target (local or remote), enough space?
• how big are you VMs?
• etc.

Tom,

I just installed ProxMox 1.7. It seems the memory overcommit function for KVM is not there anymore. Did something changed? When I create KVM VM (both windows and Linux), it grab the exact amount of memory and locked it for the particular VM. In Promox v1.5, I was able to do memory overcommitment. Could you please clarify? Thank you.

Best regards,

Andrew