| 1 |
On Wednesday, June 17, 2020 7:42:30 AM CEST n952162 wrote: |
| 2 |
> On 06/17/20 06:48, J. Roeleveld wrote: |
| 3 |
> > On Tuesday, June 16, 2020 11:08:23 PM CEST n952162 wrote: |
| 4 |
> >> On 06/16/20 22:36, J. Roeleveld wrote: |
| 5 |
|
| 6 |
<snipped> |
| 7 |
|
| 8 |
> > I have not come across MS HyperV outside of small businesses that need |
| 9 |
> > some |
| 10 |
> > local VMs. These companies tend to put all their infrastructure with one |
| 11 |
> > of |
| 12 |
> > the big cloud-VM providers (Like AWS, Azure, Googles,...) |
| 13 |
> > |
| 14 |
> > -- |
| 15 |
> > Joost |
| 16 |
> |
| 17 |
> Thank you for this excellent survey/summary. It tells me that vbox is |
| 18 |
> good for my current usages, but I should start exposing myself to Xen as |
| 19 |
> a possible migration path. |
| 20 |
|
| 21 |
I would actually suggest to read up on both Xen and KVM and try both on spare |
| 22 |
machines. |
| 23 |
See which best fits your requirements and also see if the existing management |
| 24 |
tools actually do things in a way that you can work with. |
| 25 |
|
| 26 |
My systems have evolved over the past 25-odd years and I started using Xen to |
| 27 |
reduce the amount of physical systems I had running. At the time, VMWare was |
| 28 |
expensive, KVM didn't exist yet and was missing some important features for a |
| 29 |
few years after it appeared (not sure if this exists yet, not found anything |
| 30 |
about it on KVM): |
| 31 |
- limit memory footprint of host-VM during boot. |
| 32 |
- Dedicate CPU-core(s) to the host |
| 33 |
|
| 34 |
Limiting the memory size is important, because there are several parts of the |
| 35 |
kernel (and userspace) that base their memory-settings on this amount. This is |
| 36 |
really noticable when the host thinks it has 384GB available when 370GB is |
| 37 |
passed to VMs. |
| 38 |
|
| 39 |
Dedicating CPU-cores exclusively to the host means the host will always have |
| 40 |
CPU-resources available. This is necessary because all the context-switching |
| 41 |
is handled by the host and if this stalls, the whole environment is impacted. |
| 42 |
|
| 43 |
For a lab-system, I was also missing the ability to save the full state of a |
| 44 |
VM for a snapshot. All the howto's and guides I can find online only talk |
| 45 |
about making a snapshot of the disks. Not of the memory as well. Especially |
| 46 |
when used to Virtualbox, you will notice this issue. When only snapshotting |
| 47 |
the disk, your snapshot is basically the state of when you literally pulled |
| 48 |
the plug of your VM if you want to restore back to this. |
| 49 |
|
| 50 |
For KVM, I have found a few hints that this was planned. But I have not found |
| 51 |
anything about this. Virt-manager does not (last time I looked) support Xen's |
| 52 |
functionality of storing the memory when creating snapshots either. Which is |
| 53 |
why I don't use that even for my lab/testing-server. |
| 54 |
|
| 55 |
As for tips/tricks (below works for Xen, but should also work with KVM): |
| 56 |
|
| 57 |
The way I create a new Gentoo-VM is simply to create a new block-device |
| 58 |
(Either LVM or ZFS), do all the initial steps in the chroot from the host and |
| 59 |
when it comes to the first-reboot, umount the filesystems, hook it up to a new |
| 60 |
VM and start that. |
| 61 |
|
| 62 |
Because of this, I can update the host as follows: |
| 63 |
- create new "partitions" for the host-system. |
| 64 |
- Install the latest versions, migrate the config across |
| 65 |
- reboot into the new host. |
| 66 |
|
| 67 |
If all goes fine, I can clean up the "old" partitions and prepare them for |
| 68 |
next time. If there are issues, I have a working "old" version I can quickly |
| 69 |
revert to. |
| 70 |
|
| 71 |
-- |
| 72 |
Joost |
Archives