FT> I guess that dbench generates relatively large size I/Os than FT> common workloads, which is more sensitive to latency. Right, that is what I am thinking. However, dbench inside a domU does not perform well either. Even running dbench inside a domU, the batches that the userspace daemon sees are only about 10 requests, compared to several thousand at times when running on bare metal. FT> Seems that Xen camp is pretty happy with blktap. So dm-userspace FT> should be ok with domU. Maybe it's time to try ring buffer stuff FT> again. :) You think that the ring buffer has the potential to be lower latency due to fewer memcpy()'s right? -- Dan Smith IBM Linux Technology Center Open Hypervisor Team email: danms@xxxxxxxxxx
Attachment:
pgp221Gs7ZnpR.pgp
Description: PGP signature
-- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
