Hi All,
Now i started using vhost-user and observing good performance.
Packet loss is still happening, but is around 0.006%.
[root@warmtongue ~]# ovs-ofctl dump-ports ovs
OFPST_PORT reply (xid=0x2): 6 ports
port 4: rx pkts=24717275, bytes=?, drop=?, errs=?, frame=?, over=?, crc=?
tx pkts=24727878, bytes=?, drop=1464, errs=?, coll=?
port LOCAL: rx pkts=9, bytes=662, drop=0, errs=0, frame=0, over=0, crc=0
tx pkts=1312, bytes=64834, drop=702, errs=0, coll=0
port 1: rx pkts=49454613, bytes=10789171299, drop=0, errs=0, frame=0,
over=0, crc=0
tx pkts=49426972, bytes=10787323897, drop=1961, errs=0, coll=0
port 5: rx pkts=24710294, bytes=?, drop=?, errs=?, frame=?, over=?, crc=?
tx pkts=24725186, bytes=?, drop=1603, errs=?, coll=?
port 2: rx pkts=75, bytes=?, drop=?, errs=?, frame=?, over=?, crc=?
tx pkts=1418, bytes=?, drop=10, errs=?, coll=?
port 3: rx pkts=1280, bytes=?, drop=?, errs=?, frame=?, over=?, crc=?
tx pkts=27, bytes=?, drop=10, errs=?, coll=?
[root@warmtongue ~]# ifconfig ovs
ovs: flags=323<UP,BROADCAST,RUNNING,PROMISC> mtu 1500
inet 10.54.216.88 netmask 255.255.255.0 broadcast 10.54.216.255
inet6 fe80::a236:9fff:fe06:5e08 prefixlen 64 scopeid 0x20<link>
ether a0:36:9f:06:5e:08 txqueuelen 500 (Ethernet)
RX packets 1315 bytes 64984 (63.4 KiB)
RX errors 0 dropped 704 overruns 0 frame 0
TX packets 9 bytes 662 (662.0 B)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
On Mon, Jul 6, 2015 at 7:00 PM, Gray, Mark D <mark.d.g...@intel.com> wrote:
> >
> > I'm sending traffic to a macvtap2 interface on KVM instance [derived from
> > pktovs bridge] and the traffic is voice [udp data ~104 bytes with ip
> header
> > and payload]
> >
> > Following are the interface details [you can see the pktovs dropped value
> > also]
> >
> > macvtap0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
> > macvtap1: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
> >
> >
> > macvtap2: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
> >
> > inet6 fe80::5054:ff:fed1:ad9c prefixlen 64 scopeid 0x20<link>
> > ether 52:54:00:d1:ad:9c txqueuelen 500 (Ethernet)
> > RX packets 277957152 bytes 18691630130 (17.4 GiB)
> > RX errors 6226 dropped 6226 overruns 0 frame 0
> > TX packets 5913091 bytes 1270520006 (1.1 GiB)
> > TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
> >
> > macvtap3: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
> > inet6 fe80::5054:ff:fec6:8151 prefixlen 64 scopeid 0x20<link>
> > ether 52:54:00:c6:81:51 txqueuelen 500 (Ethernet)
> > RX packets 14488698 bytes 3102135336 (2.8 GiB)
> > RX errors 0 dropped 0 overruns 0 frame 0
> > TX packets 6403020 bytes 1386693403 (1.2 GiB)
> > TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
> >
> > pktovs: flags=323<UP,BROADCAST,RUNNING,PROMISC> mtu 1500
> >
> > inet 10.54.218.81 netmask 255.255.255.0 broadcast 10.54.218.255
> > inet6 fe80::e611:5bff:fe98:993 prefixlen 64 scopeid 0x20<link>
> > ether e4:11:5b:98:09:93 txqueuelen 4096 (Ethernet)
> > RX packets 146219247 bytes 10896680781 (10.1 GiB)
> > RX errors 0 dropped 5215 overruns 0 frame 0
> > TX packets 5620563 bytes 1220607110 (1.1 GiB)
> > TX errors 0 dropped 6695584 overruns 0 carrier 0 collisions 0
> >
> >
> As I mentioned in a previous post, you need to use vhost-user. The
> bottleneck
> you are experiencing is the tap device. vhost-user will give you better
> performance
> into the virtual machine, especially if you use a DPDK poll mode driver in
> the guest.
>
> http://openvswitch.org/support/ovscon2014/18/1530-dpdk-accelerating.pptx
>
> https://www.youtube.com/watch?v=FmQmo8zG4y8&list=PL9MeVsU0uG65Ne9_qJNi4J8AzXvP05gjZ&index=23
>
> https://github.com/openvswitch/ovs/blob/master/INSTALL.DPDK.md#dpdk-vhost
>
>
>
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
