Re: Virtio drivers (virtio-net)

[joris dedieu <joris.dedieu@xxxxxxxxx>]

2011/1/23 Tim Bisson <bissont@mac.com>:
> Hi All,
>
> There's been some activity in the FreeBSD mailing lists about the virtio
> drivers. A couple of people presented their drivers:
> http://lists.freebsd.org/pipermail/freebsd-current/2011-January/022036.html
> http://lists.freebsd.org/pipermail/freebsd-current/2011-January/022045.html
>
> We took a look at these drivers and found that the first link seemed to be a
> really clean implementation of the virtio infrastructure and virtio-net
> driver. We decided to stop the porting effort of the NetBSD-based virtio
> drivers and instead use the FreeBSD ones. We successfully ported the FreeBSD
> virtio infrastructure and virtio-net driver, which now works and is ready
> for more testers.
>
> In order to port this driver, we had to make some changes to the kernel:
> essentially add kern/subr_sglist.c from FreeBSD and make a probe interface
> public. The patch file is in the top-level directory of the git tree and was
> diff'd from master this morning. The file name is:
> 0001-kern-changes-for-freeebsd-virtio-driver-port.patch
>
> Regarding the original virtio-blk driver we ported from NetBSD. I will
> probably port this to the new virtio-infrastructure, unless the FreeBSD
> virtio-blk driver is posted on the FreeBSD mailing list in the near future.
>
> Here's a link to the code:
> http://gitorious.org/dbd-virtio-drivers-freebsd-port
>
> and some notes:
> http://gitorious.org/dbd-virtio-drivers-freebsd-port/pages/Home
>
> Regarding performance, I don't have a good explanation yet, but emulation
> with the re driver is actually faster than virtio on a simle scp test:
>
I think that you may not focus on such a test. IMHO virtio goal is to
be more scalable than pure emulation. Run a test with iperf or
bonnie++ on serval vm in the same host. It should sound very different
with or without virtio. Without you should see the vms working hard,
with you should see more work on the host itself.

Cheers
Joris
> re driver
> $ time scp 192.168.122.140:200m .
> 200m                       100%  200MB  15.4MB/s   00:13
>
> virtio driver
> $ time scp 192.168.123.8:200m .
> 200m                       100%  200MB  11.8MB/s   00:17
>
> ifconfig from dbsd vm for above test
> $ ifconfig
> re0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
>    options=1b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING>
>    inet6 fe80::5054:ff:fe03:9900%re0 prefixlen 64 scopeid 0x1
>    inet 192.168.122.140 netmask 0xffffff00 broadcast 192.168.122.255
>    ether 52:54:00:03:99:00
>    media: Ethernet autoselect (100baseTX <full-duplex>)
>    status: active
> faith0: flags=8002<BROADCAST,MULTICAST> mtu 1500
> lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> mtu 16384
>    options=3<RXCSUM,TXCSUM>
>    inet 127.0.0.1 netmask 0xff000000
>    inet6 ::1 prefixlen 128
>    inet6 fe80::1%lo0 prefixlen 64 scopeid 0x3
> ppp0: flags=8010<POINTOPOINT,MULTICAST> mtu 1500
> sl0: flags=c010<POINTOPOINT,LINK2,MULTICAST> mtu 552
> vtnet0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 1500
>    options=28<VLAN_MTU,JUMBO_MTU>
>    inet6 fe80::5054:ff:fe33:3bd3%vtnet0 prefixlen 64 scopeid 0x6
>    inet 192.168.123.8 netmask 0xffffff00 broadcast 192.168.123.255
>    ether 52:54:00:33:3b:d3
>
> Comments and suggestions are always welcome.
> Tim
>