Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hi all, I am again testing drbd for use with xen virtual machines. On a typical physical pair of machines, there will be about 8 to 15 (maybe even 20) virtual machines. Each VM needs its own block device for storage. To be able to fail over or migrate single VMs between the nodes, I must be able independently activate the block devices on one of the two nodes. That raises the question of how many drbd devices one can reasonably use on a single cluster of two nodes without too much overhead. I just guess that in-kernel locking and synchronization, and the number of kernel threads could be an issue with a higher number of devices (I am no expert here). Are 15 to 20 devices a reasonable number? There are other options. For example drbd v0.8/8.0 (?) with primary/primary in combination with CLVM [1] would allow for a single shared drbd device with logical volumes for the VMs. Of course, a single logical volume would be accessed from only one host at a time, but the drbd device itself would be active on both nodes. Does anybody have experience with DRBD and RedHat Cluster Suite integration? Another option, if DRBD and CS do not play well together, would be a custom frontend for device-mapper to create logical volumes, but integrated with heartbeat and more aware of the underlying drbd. So the basic question is still how many drbd devices one can reasonably use. Best Regards, Michael Paesold [1] Cluster LVM; a cluster-aware metadata management for LVM, part of RedHat GFS/Cluster Suite