netdev - KVM vs Xen-PV netperf numbers

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Date:	Fri, 5 Dec 2014 12:02:18 -0800
From:	Nick H <nickkvm@...il.com>
To:	netdev@...r.kernel.org
Subject: KVM vs Xen-PV netperf numbers

Hello

Not sure I have the right audience, I have two VM's on similar hosts.
One VM is KVM + virtio + vhost based while other is Xen PV with xen
netfront driver. Running a simple netperf test for 1400 byte packets
on both VM's, I see wider difference between throughput as follows.

Xen-pv throughput comes to :

./netperf -H 10.xx.xx.49 -l 20 -t UDP_STREAM -- -m 1400
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
10.xx.xx.49 (10.xx.xx.49) port 0 AF_INET : demo
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992    1400   20.00     1910549      0    1069.89
262144           20.00     1704789            954.66

whereas KVM virtio number comes to :

./netperf -t UDP_STREAM -l 10 -H 10.xx.xx.49 -- -m 1400
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
10.xx.xx.49 (10.xx.xx.49) port 0 AF_INET : demo
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992    1400   10.00      155060      0     173.65
262144           10.00      155060            173.65

I built a custom kernel where I simply free up the (UDP only) skb in
virtio: xmit_skb() routine and I count how many skb's I have received.
Surprisingly it was not too high either:

./netperf -t UDP_STREAM -l 10 -H 10.xx.xx.49 -- -m 1400
MIGRATED UDP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
10.xx.xx.49 (10.xx.xx.49) port 0 AF_INET : demo
Socket  Message  Elapsed      Messages
Size    Size     Time         Okay Errors   Throughput
bytes   bytes    secs            #      #   10^6bits/sec

212992    1400   10.00      224792      0     251.74
262144           10.00           0              0.00

1910549 packets pumped in xen pv driver versus 212992 packets pumped
in case of virtio driver. Assuming the data path inside the kernel is
same for both drivers , and I have eliminated virtio's
virtqueue_kick() call by freeing the packet ahead in my experiment,
can all this overhead attributed to system call overhead in case of
KVM+virtio combination ? Anything I am missing ?

The KVM setup is based off:
Linux ubn-nested 3.17.0+ #16 SMP Thu Dec 4 12:00:09 PST 2014 x86_64
x86_64 x86_64 GNU/Linux

Regards
N
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html