Hi Fujita, I've been very pleased with the removal of the remap caching from dm-userspace because performance seems to be good and complexity is lower. However, all of my tests up until now have been with dbench. I think that dbench works well in cases where latency is high, which is of course the case since we're sending all requests to userspace. When I try to boot a Xen domU with root on dm-userspace (without the caching), it is *painfully* slow. I assume this is because the domU boot is doing a lot of synchronous, sequential operations. Since domU performance was good when I was caching mappings in the kernel, I would assume that the difference in performance is related to the caching. What do you think? -- Dan Smith IBM Linux Technology Center Open Hypervisor Team email: danms us ibm com
Attachment:
pgpz9oaHqQrP2.pgp
Description: PGP signature