vhost-blk
- started w/ vhost-net, nice and modular
- qemu merging requests, so outperforming (throughput) for sequential write
- random read/write and sequential reads are comparable or better
- can't do e.g. qcow2
- spreading work across all cpu workqueues
- are there cases where we expect vhost-blk is needed?
- not driven by poor perf benchmarks, more driven from interest/curiosity
- hch's numbers (esp. large request) are solid
- low hanging fruit...no interrupt mitigation, addr lookup on all request
- get some performance numbers and profiles to see what we can do to
improve existing virtio before moving to vhost
qemu optimizations (re: above addr lookup)
- no long lived ptr to guest memory
- ram memory patch allowed for stable mapping
- objection was memory hotplug/remove
- how much do we want to gate qemu progress on memory hotplug?
- http://wiki.qemu.org/Features/RamAPI
ide emulation
- qemu virtio-blk vs. other hv ide emulation...ide was better
- this was some time ago
- older qemu, even qemu ide emulation was better than virtio-blk in
some cases
- ide still bounce buffering, and has some fundamental limitations
(some of those may be fixable w/ ahci)
- anyone looked at this recently?
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
