All of lore.kernel.org
 help / color / mirror / Atom feed
* [Qemu-devel] 9p performance
@ 2016-09-06 14:21 Chao Peng
  0 siblings, 0 replies; only message in thread
From: Chao Peng @ 2016-09-06 14:21 UTC (permalink / raw)
  To: qemu-devel

Hi,

I had some initial investigation on 9p read/write performance and it
looks not good when compared to the implementation in kvmtool.

With kvmtool, the write speed is ~1400MB/s. With QEMU, it's only ~
700MB/s.

I had a quick look at the code, the kvmtool code is really simple so no
other overhead exists. For QEMU, it's different. There are some
indirect layers on the path from guest call to the final 9p driver (in
our case it's local).

The biggest overhead I have seen is the coroutine framework 9p replies
on. I tried to disable coroutine for 9p temporarily and traced it, the
time in v9fs_write decreased from ~60us -> ~30us, at the same time the
write speed in guest increased from ~700MB/s -> ~1100MB/s.

There is still another ~300MB/s needs to be figured out when compared
to kvmtool.

Anyone have experience on this or any suggestions?

Thanks,
Chao

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2016-09-06 14:21 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-09-06 14:21 [Qemu-devel] 9p performance Chao Peng

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.