1 |
On Wednesday, December 31, 2014 09:42:11 AM Rich Freeman wrote: |
2 |
> On Wed, Dec 31, 2014 at 8:26 AM, Mick <michaelkintzios@×××××.com> wrote: |
3 |
> > On Wednesday 31 Dec 2014 12:47:55 Sid S wrote: |
4 |
> > |
5 |
> > Vbox seems to be coming last by quite some margin in the intel tests! I |
6 |
> > also read this article and it looks that vbox is thankfully doing better |
7 |
> > on AMD; but there are differences in the versions and kernels used |
8 |
> > between the two |
9 |
> > articles: |
10 |
> I think you need to think about your use case. The requirements were |
11 |
> for a workstation testing environment. I think performance (as long |
12 |
> as somewhat reasonable) isn't going to be a big concern there vs ease |
13 |
> of setup, ability to snapshot, |
14 |
|
15 |
The thing lacking from KVM (and I believe also Containers) is that the memory |
16 |
contents are not included in snapshots. Making the snapshots basically result |
17 |
in an unclean-shutdown scenario. |
18 |
Which is ok-ish as a backup, but not when testing different steps where a quick |
19 |
and easy roll-back is often required. |
20 |
|
21 |
> convenience features like being able to |
22 |
> group guests, being able to get the right environment easily, etc. |
23 |
> You probably also want reasonable graphics performance if you're |
24 |
> testing clients inside VMs. If performance makes the difference |
25 |
> between being able to run the cluster you need to test on your |
26 |
> workstation or not, then that becomes a factor. Otherwise it is a |
27 |
> nice-to-have. |
28 |
> |
29 |
> If you're talking about running servers then performance becomes much |
30 |
> more important. However, if you're running linux guests you should |
31 |
> seriously consider containers, and if containers aren't the right |
32 |
> solution you should also be looking at stuff like VMWare (I don't know |
33 |
> how well the FOSS solutions do as far as enterprise-y features go). |
34 |
|
35 |
I compared the ease-of-use and performance between XenServer, VMWare and |
36 |
VirtualBox. |
37 |
VMWare generally is the slower of the three. |
38 |
Also, the weird errors occuring when VMs are migrated between nodes in a |
39 |
VMWare cluster makes me worry every time I hear it's being used for critical |
40 |
systems. |
41 |
|
42 |
> In any case, while not quite as simple as Virtualbox I've found that |
43 |
> virt-manager is very easy to use once you've gotten networking set up |
44 |
> (which isn't too hard to do under either openrc or networkd). I tend |
45 |
> to use the GUI for setting things up and for graphical guests, and I |
46 |
> used to create init.d scripts / units for the stuff that I |
47 |
> subsequently moved to containers. You can go back-and-forth between |
48 |
> the two (and to be fair you can do the same with virtualbox). One of |
49 |
> the advantages of KVM is that it doesn't require tainting your kernel, |
50 |
|
51 |
That is an advantage of KVM and Xen over Virtualbox and VMWare. |
52 |
|
53 |
> and you don't have to remember to rebuild the module anytime you |
54 |
> update your kernel. I've finally gotten to the point where I don't |
55 |
> have any external modules on one of my boxes and I'm very happy with |
56 |
> that (alas, my mythtv frontend needs nvidia-drivers - I don't think |
57 |
> the hardware acceleration is as good with the kernel drivers though to |
58 |
> be fair it has been a year or two since I last tried). |
59 |
|
60 |
I tend to use the nvidia-drivers where I need graphics. But those machines are |
61 |
not VMs. |
62 |
If graphical performance is a requirement, NVidia cards (apart from the |
63 |
expensive professional ones) are best avoided. They are actively crippled in a |
64 |
VM environment. |
65 |
|
66 |
-- |
67 |
Joost |