Hello, Am 25.03.2010 um 13:49 Uhr schrieb Pasi Kärkkäinen <pasik@iki.fi>: > On Tue, Mar 23, 2010 at 03:58:57PM +0100, Markus Hochholdinger wrote: [..] > Btw what kind of shared storage are you using? I use lvm2 to use the hard disks of one server, cut out logical volumes with stripping over all hard disks for maximum performance. Each logical volume is exportet via iSCSI to the other servers. For one domU i use two logical volumes from different (storage-)servers in a raid1 array INSIDE the domU. Notes: * I do NOT reimport exported iSCSI LUNs on the same host! Instead i use the logical volumes directly. * I made scripts which make a consistent nameing accross all servers. This scripts create links in /etc/xbd/ of the form $DOMUNAME.$STORAGEHOST . So on each xen host the path to the same logical volume is always the same! This is very important for live migration! * I don't use any caching of the block devices! It's better for performance AND there could be data inconsistency if you use any caching while doing live migration! * I don't nead heartbeat to have fully redundant storage! * There can be NO SPLIT BRAIN!!! * Each server in my setup ist a xen host and also a storage host. * My domUs are all linux, i use software raid (md) with bitmaps, so i can do rebuild very fast if one storage host was short offline (e.g. for maintenance). * It's ALL within Debian lenny, except my scripts! * I can use ONE hardware as xen host AND as storage host! * My cluster setup scales well from 2 to n. * I'm doing this setup since april 2006, but in the beginning i used gnbd instead of iscsi. So changing technologies is very easy with this setup! Drawback: I have to manage the raid1 inside each domU! But it's no problem if the admin of the hardware is also the admin of the domUs. If you are interested I could give more details, just ask. -- greetings eMHa
Attachment:
signature.asc
Description: This is a digitally signed message part.