Putting the image on tmpfs cut the snapshot time down by quite a bit. It didn’t do much for the restore time, probably because the images I’m using fit in the file cache already.
Now I’m going to look in to profiling qemu to see where it’s spending all of its time.
From: Doug Hughes <doug.hughes@xxxxxxxxxxxxxxx>
Date: Tuesday, September 19, 2017 at 12:59 PM
To: "Jackson, Gary L." <Gary.Jackson@xxxxxxxxxx>, "libvirt-users@xxxxxxxxxx" <libvirt-users@xxxxxxxxxx>
Subject: Re: [libvirt-users] Rollback to running VM
I would estimate you are bounded by memory bandwidth (GB/sec - so not much, but proportional to memory size) and disk bandwidth. So, using SSD will help for certain vs spinning. I'm not sure about the catch-22 implicit with using a ramdisk, but since the
ramdisk should be previsioned from the host to the guest, it seems like it should work.
There may be other bottlenecks in drivers, etc. Not in my knowledge domain.
On 9/19/2017 12:53 PM, Jackson, Gary L. wrote:
Thanks! The vanilla snapshot-create-as and revert commands work just exactly the way I’d want them to. I’d gotten a bit confused by the documentation and went down a rabbit hole with external snapshots.
Is there any way to make the snapshot and restore faster, like maybe by a decimal order of magnitude? For example, will PXE booting the VM and running it entirely off of a RAM disk and also using a RAM disk for the backing store help?
According to this page:
https://wiki.libvirt.org/page/Snapshots
It saves complete memory state, or crash-consistent disk state, but not both.
You probably want to test it thoroughly and see how it does. I only use the disk-consistent snapshots.
On 9/19/2017 11:17 AM, Jackson, Gary L. wrote:
Will snapshot-revert restore the processor and memory state of the VM as it was at the time of the snapshot?
--
|
Doug Hughes
Keystone NAP
Fairless Hills, PA
1.844.KEYBLOCK (539.2562)
|
|
--
|
Doug Hughes
Keystone NAP
Fairless Hills, PA
1.844.KEYBLOCK (539.2562)
|
|
