The reason I want to see all cores in dom0 is that I am developing a new batch scheduler for a cluster,
and in this situation, I need to use real and virtual cluster together.
Due to performance degradation in virtual cluster, I can not run all parallel job on virtual machines.
So the idea is just to run parallel jobs in dom0 and use virtual machine whenever is necessary to get better performance.
That's why i want to see all cores and RAM in dom0.
I think I know the difference between xen hypervisor and centos dom0 kernel.
As I know, Xen 3.x let dom0 kernel to see only 32 cores and 32GB RAM, but in Xen 4.x. it can be upgraded to more CPU and RAM.
The question I have is why Xen panics during
boot?!
------------------------------------------------------------------
------------------------------------------------------------------
> Hi
>
> Has anybody got an idea why xen 4.0.1 panics during boot?!
> I have a Tyan server with 48 AMD cores and 128GB RAM. I compiled and
> installed xen 3.X along with CentOS 5.5 default xen kernel. Everything was
> OK there. The only problem I had was that CentOS kernel is allowed to see
> only 32 cores and 32GB of RAM which is actually supposed to be. However I
> wanted to see all cores and RAM in dom0,
Why?
For most practical purposes, it's useless to have all cores and RAM in dom0:
- you'd want to start domU's, which would require memory, thus dom0
memory would need to
balloon-down. Doing so involves some overhead
(and depending on some circumstances like the amount ballooned-down,
it can cause crash)
- domUs will need some processing power, thus making all cores
available to dom0 is kinda pointless as domU will take some processing
power from it anyway
Best practice is to:
- dedicate a certain amount of RAM to dom0 (usually 512M - 1G is
enough), so dom0 will never need to balloon down
- dedicate one core (or more, depending on what you run on dom0) and
set domUs to use other cores.
> so I went for xen 4.0.1
Why?
Do you know the difference between xen HYPERVISOR and dom0 KERNEL?
If you say "CentOS kernel is allowed to see only 32 cores and 32GB of
RAM" then you should be compiling a kernel, not the hypervisor.
FWIW, if you need to use Xen 4.0.1 hypervisor with RHEL/Centos
kernel-xen, last time I check yo need "pci=nomsi" on
some
configurations (in my case it was HP smart array). Also, comment-out
"quiet" and "rhgb" so you can see better what cause the kernel panic.
Your grub/menu.lst should look something like this
title CentOS (2.6.18-194.el5.customxen)
root (hd0,1)
kernel /boot/xen-4.0.1.gz
module /boot/vmlinuz-2.6.18-194.el5.customxen ro root=LABEL=/ pci=nomsi
module /boot/initrd-2.6.18-194.el5.customxen.img
--
Fajar