[Sheepdog] [PATCH 0/2] use corosync for the cluster communication
morita.kazutaka at lab.ntt.co.jp
Thu Nov 12 21:45:07 CET 2009
This is another implementation of `dog' (cluster manager daemon) in C language.
This patch don't support full feature of dog. I named this tiny program `puppy'.
- create VM image
- snapshot VM image
- clone VM image
- detect a newly added machine automatically
- scale to 128 nodes (maybe)
- get rid of JVM dependency of Sheepdog
- recover data objects in any case
- add a new machine safely (existing data may be lost)
- shutdown Sheepdog cluster safely
- scale to several hundreds nodes
If you don't change the number of machines after mkfs, puppy should work well.
Sheepdog can run with only one machine if you don't care about data redundancy.
I hope more people try to test.
My test environment is debian lenny amd64 and corosync 1.1.2.
Before running sheep and puppy, don't forget to start corosync daemon.
More information about the sheepdog