[sheepdog-users] 答复:sheepdog data write performance with 5 node

[Liu Yuan]

On Mon, Nov 25, 2013 at 09:06:55AM +0100, Gerald Richter - ECOS wrote:
> Thanks, one more question:
> 
> Does the qemu cache setting (i.e. cache=none,writeback,unsafe,directsync,writethrough) has any impact on sheepdog cache handling and the usage of the object cache?
> 

Yes, actually cache=type is the only means to control what kind of cache it is
for specific VM. This means our object cache is set on VM basis, even though
it is a shared cache by all VMs on this node. (By 'share', we indeed share read
only objects between VMs to save space)

	QEMU cache=     |   sheep with object cache enabled
	none,directync  |   bypass object cache
	writeback       |   writeback mode, that is writeback on sync request of VM
	writethrough    |   writethrough mode, every write is assured to reach backend
	unsafe          |   writeback mode, but VM won't issue sync at all. Purge and
	                    write the data back when the VM is released.

There is one more thing to notice, for object cache, sheepdog have a background
flusher invisible to VM which do write back of dirty data if some dirty ratio is
reached to boost 'sync' performance like how linux memory manager works on page
cache.

Thanks
Yuan