[sheepdog] sheepdog vs raid 5/6

Liu Yuan namei.unix at gmail.com
Thu Jan 23 09:59:47 CET 2014


On Thu, Jan 23, 2014 at 12:24:52PM +0400, Vasiliy Tolstov wrote:
> Hi all. If i don't need cluster storage for vps now, but i want to
> have hdd fail protection, what is the best to use for storage for vm?
> I have 5 1Tb drives and can create raid 6 and use lvm for it, or
> format each drive and use it in sheepdog (qemu connects to local
> sheepdog daemon). in case of erasure coding - does it possible to lost
> 2 drives and continue to work?
> How about write/read speed of this ?
> Does somebody try this?

This is absolutely feasible with sheepdog.

follow my instructions:

1. git clone https://github.com/sheepdog/sheepdog.git
2. ./autogen.sh;./configure --disable-corosync;make install
3.
  for i in `seq 1 5`; do
  $ sheep -n -z $i -c local disk$i # disk$i is mountponit of your disks
  done
4. dog cluster format -c 2:3

With above setup, you actually create 3 parities for data, meaning that you can
survive 3 disks failure at the same time! The redundancy is 1.5x.

Okay, then enjoy. For performance, you can test on your own. I'd suggest you use
virtio for sheepdog volume and use new QEMU like v1.7

For more infomation abbout EC in sheepdog, please see

https://github.com/sheepdog/sheepdog/wiki/Erasure-Code-Support

Thanks
Yuan



More information about the sheepdog mailing list