2011/2/14 Fajar A. Nugraha
<list@xxxxxxxxx>
On Tue, Feb 15, 2011 at 3:41 AM,
lucianobarreto@xxxxxxxxx
<
lucianobarreto@xxxxxxxxx> wrote:
> Sorry if duplicated...
> Thanks guys, let me expose my problem. I'm prototyping a fault tolerance
> server (byzantine fault tolerance). Vms need do comunicate each order for
> make decisions about requests. My host will be out of network (cant access
> anything) and my Vms can comunicate to the external world, but the decision
> need to be local.
Most HA-systems are designed to work whether it's on a VM or on a
physical machine. Normally you'd just use whatever mechanism you'd use
on physical machine/
It's an academical model that I'm trying to propose to my master thesis. On general models physical machines or vm exchange informations over the network (point of fail, man-in-the-middle, non reliable channels), than my propose is to use another way to exchange this informations.
> Yes, I could comunicate via network inside VM to exchange
> information but my propose is to have a reliable local share to do it. In
> another work a friend used a Virtualbox folder share to do it, but now I
> need increase performance and use Xen to do it.
Virtualbox shared folder has about similar performance as running
nfs/samba inside dom0 and mount it on domU.
Yes, i need more performance not just in exchange information, but running the VMs (my choice to Xen).
> About clustered filesystems, any options to use it in a normal block device
> like Sata HD or SCSI? I understand..."clustered"... but any option?
> Thanks and I will get more information about clustered filesystems
You should be able to force-assign a block device to more than one
domUs with "w!" instead of just the normal "w" (e.g.
phy:/dev/sda1,xvda,w!), skipping the normal
you-can-only-assign-a-block-device-read-write-once check. Then you can
run a cluster file system (OCFS2/GFS2/whatever) on top of it.
I'll try it....
IMHO it's too complicated for your purpose though. Better stick to
nfs/smb on dom0.
--
Fajar
