Linux Virtualization - Aug 2009 - vhost net: performance with ping benchmark

At Rusty's suggestion, I tested vhost base performance with ping.
Results below, and seem to be what you'd expect.  I'm working on TSO
support, expect results shortly.



latency with ping (lower is better):
native:
[root at virtlab17 ~]# ping -c 1000000 -f -q  21.1.50.4
PING 21.1.50.4 (21.1.50.4) 56(84) bytes of data.

--- 21.1.50.4 ping statistics ---
1000000 packets transmitted, 1000000 received, 0% packet loss, time 73624ms
rtt min/avg/max/mdev = 0.047/0.061/1.253/0.036 ms, ipg/ewma 0.073/0.097 ms

vhost:
[root at virtlab17 ~]# ping -c 1000000 -f -q  20.1.50.4
PING 20.1.50.4 (20.1.50.4) 56(84) bytes of data.

--- 20.1.50.4 ping statistics ---
1000000 packets transmitted, 1000000 received, 0% packet loss, time 92308ms
rtt min/avg/max/mdev = 0.064/0.080/1.062/0.041 ms, ipg/ewma 0.092/0.083 ms

userspace:
[root at virtlab17 ~]# ping -c 100000 -f -q  20.1.50.4
PING 20.1.50.4 (20.1.50.4) 56(84) bytes of data.

--- 20.1.50.4 ping statistics ---
100000 packets transmitted, 100000 received, 0% packet loss, time 54473ms
rtt min/avg/max/mdev = 0.219/0.505/2.342/0.131 ms, ipg/ewma 0.544/0.485 ms


Conclusion: for latency difference between native and vhost is about
20usec, userspace is way slower.
This basically matches what was observed with venet.

##

throughput with ping (lower time is better):

native:
[root at virtlab17 ~]# ping -s 1024 -l 120 -c 100000 -f -q 20.1.50.2
PING 20.1.50.2 (20.1.50.2) 1024(1052) bytes of data.

--- 20.1.50.2 ping statistics ---
100000 packets transmitted, 100000 received, 0% packet loss, time 3582ms
rtt min/avg/max/mdev = 0.105/4.155/5.471/0.471 ms, pipe 120, ipg/ewma
0.035/4.567 ms

vhost:
[root at virtlab17 ~]# ping -s 1024 -l 120 -c 100000 -f -q 20.1.50.4
PING 20.1.50.4 (20.1.50.4) 1024(1052) bytes of data.

--- 20.1.50.4 ping statistics ---
100000 packets transmitted, 100000 received, 0% packet loss, time 3900ms
rtt min/avg/max/mdev = 0.354/4.129/6.009/0.520 ms, pipe 120, ipg/ewma
0.039/3.109 ms

userspace:

[root at virtlab17 ~]# ping -s 1024 -l 120 -c 1000000 -f -q 20.1.50.4
PING 20.1.50.4 (20.1.50.4) 1024(1052) bytes of data.

--- 20.1.50.4 ping statistics ---
1000000 packets transmitted, 999731 received, 0% packet loss, time 45082ms
rtt min/avg/max/mdev = 0.299/4.130/8.143/1.094 ms, pipe 120, ipg/ewma
0.045/1.117 ms

Conclusion: for throughput vhost is half-way between native and
userspace. Again, same thing as was observed with venet.

-- 
MST

On Mon, Aug 24, 2009 at 11:12:41AM +0300, Michael S. Tsirkin
wrote:
> At Rusty's suggestion, I tested vhost base performance with ping.
> Results below, and seem to be what you'd expect.
Rusty, any chance you could look at the code?  Is it in reasonable
shape? I think it makes sense to merge it through you. What do you
think?  One comment on file placement: I put files under a separate
vhost directory to avoid confusion with virtio-net which runs in guest.
Does this sound sane?  Also, can a minimal version (without TSO, tap or
any other features) be merged upstream first so that features can be
added later? Or do we have to wait until it's more full featured?
Finally, can it reasonably make 2.6.32, or you think it needs more time
out of tree?

Thanks very much,

-- 
MST