[ovs-discuss] throughput when using LOCAL port
Naoyuki NS Shimizu
naoyuki.ns.shimizu at nts.ricoh.co.jp
Wed Nov 30 06:36:07 UTC 2016
Hi,
"it just takes more time to process packets arriving at the local TCP/IP stack than it does to send packets to a physical port."
âWhy? I think via local TCP/IP stack is faster than via physical port.
Now when packets arrive at the local TCP/IP stack, packets just send.
-----ä½æè
: Ben Pfaff <blp at ovn.org> -----
å®å
: Naoyuki NS Shimizu/R/RICOH at RICOH
éä¿¡å
: Ben Pfaff <blp at ovn.org>
æ¥ä»: 2016/11/30 09:47AM
Cc: ovs-discuss at openvswitch.org
件å: Re: [ovs-discuss] throughput when using LOCAL port
Probably it just takes more time to process packets arriving at the
local TCP/IP stack than it does to send packets to a physical port.
What's happening to the packets when they arrive at the local TCP/IP
stack?
On Tue, Nov 29, 2016 at 10:54:04AM +0000, Naoyuki NS Shimizu wrote:
>
>
> Hi,
>
>
>
> Because I made following flow.
>
>
> cookie=0x0, duration=1883.504s, table=0, n_packets=1, n_bytes=60, idle_age=1876, priority=3,udp,tp_dst=5001 actions=mod_dl_dst:(PC3(connected to eth3) or LOCAL MAC),mod_nw_dst:(PC3(connected to eth3) or LOCAL IP),mod_tp_dst:4000,output:(eth3 or LOCAL)
> cookie=0x0, duration=6649.281s, table=0, n_packets=3, n_bytes=182, idle_age=1876, priority=4,udp,nw_src=(PC3(connected to eth3) or LOCAL IP),nw_dst=(PC2(connected to eth2) IP),tp_dst=5001 actions=mod_nw_src:(PC1(connected to eth1) IP),mod_dl_src:(PC1(connected to eth1) MAC),output: eth2
> cookie=0x0, duration=6651.852s, table=0, n_packets=1679, n_bytes=100740, idle_age=2, priority=1,arp actions=ALL
>
>
>
>
> Why I cannot get nearly 1Gbps throughput when I use LOCAL port?
>
>
> -----ä½æè
: Ben Pfaff <blp at ovn.org> -----
>
>
> å®å
: Naoyuki NS Shimizu/R/RICOH at RICOH
> éä¿¡å
: Ben Pfaff <blp at ovn.org>
> æ¥ä»: 2016/11/27 07:12AM
> Cc: ovs-discuss at openvswitch.org
> 件å: Re: [ovs-discuss] throughput when using LOCAL port
>
>
> Still doesn't make sense to me. There is no such thing as "sending
> packets from eth1 to eth2". Instead, OVS receives a packet from eth1
> and does something to it. This might be sending it to eth2 or to the
> local port. Your steps 1 and 2, therefore, don't make any sense to me.
>
> Can you explain what you're doing in terms of how OVS works?
>
> On Thu, Nov 24, 2016 at 12:47:37AM +0000, Naoyuki NS Shimizu wrote:
> > Thank you for your reply.
> >
> >
> >
> > I want to add infomation to packets in LOCAL port.
> >
> > Pakcet flow is below.
> >
> > 1. I send packets from eth1 to eth2
> >
> > 2. Switch forward packets to LOCAL port.
> >
> > 3. I add informatin to packets in LOCAL port. (Now I have not done anything)
> >
> > 4. I send packets include additional infomation from LOCAL port to eth2.
> >
> > 5. Switch forward packets include additional informatin to eth2.
> >
> > 6. I receive packets include additional information in eth2.
> >
> >
> >
> >
> > When I use eth3 instead of LOCAL port, I got nearly 1Gbps throughput.
> >
> > But when I use LOCAL port, I got nearly 100Mbps throughput.
> >
> > Why I cannot get nearly 1Gbps throughput when I use LOCAL port?
> >
> >
> > -----ä½æè
: Ben Pfaff <blp at ovn.org> -----
> >
> >
> > å®å
: Naoyuki NS Shimizu/R/RICOH at RICOH
> > éä¿¡å
: Ben Pfaff <blp at ovn.org>
> > æ¥ä»: 2016/11/23 03:25AM
> > Cc: ovs-discuss at openvswitch.org
> > 件å: Re: [ovs-discuss] throughput when using LOCAL port
> >
> >
> > On Tue, Nov 22, 2016 at 08:25:49AM +0000, Naoyuki NS Shimizu wrote:
> > > Hi, everyone.
> > >
> > > I made a switch named ofs0 and added eth1, eth2 and eth3 port.
> > >
> > > Details are below.
> > >
> > >
> > >
> > > #ovs-ofctl show ofs0
> > > OFPT_FEATURES_REPLY (xid=0x2): dpid:0000000a850722e9
> > > n_tables:254, n_buffers:256
> > > capabilities: FLOW_STATS TABLE_STATS PORT_STATS QUEUE_STATS ARP_MATCH_IP
> > > actions: OUTPUT SET_VLAN_VID SET_VLAN_PCP STRIP_VLAN SET_DL_SRC SET_DL_DST SET_NW_SRC SET_NW_DST SET_NW_TOS SET_TP_SRC SET_TP_DST ENQUEUE
> > > 1(eth1): addr:00:0a:85:07:22:e9
> > > config: 0
> > > state: 0
> > > current: 1GB-FD AUTO_NEG
> > > advertised: 10MB-HD 10MB-FD 100MB-HD 100MB-FD 1GB-HD 1GB-FD COPPER AUTO_NEG
> > > supported: 10MB-HD 10MB-FD 100MB-HD 100MB-FD 1GB-HD 1GB-FD COPPER AUTO_NEG
> > > speed: 1000 Mbps now, 1000 Mbps max
> > > 2(eth2): addr:00:0a:85:07:22:ea
> > > config: 0
> > > state: 0
> > > current: 1GB-FD AUTO_NEG
> > > advertised: 10MB-HD 10MB-FD 100MB-HD 100MB-FD 1GB-HD 1GB-FD COPPER AUTO_NEG
> > > supported: 10MB-HD 10MB-FD 100MB-HD 100MB-FD 1GB-HD 1GB-FD COPPER AUTO_NEG
> > > speed: 1000 Mbps now, 1000 Mbps max
> > >
> > > 3(eth3): addr:00:0a:85:07:22:eb
> > > config: 0
> > > state: 0
> > > current: 1GB-FD AUTO_NEG
> > > advertised: 10MB-HD 10MB-FD 100MB-HD 100MB-FD 1GB-HD 1GB-FD COPPER AUTO_NEG
> > > supported: 10MB-HD 10MB-FD 100MB-HD 100MB-FD 1GB-HD 1GB-FD COPPER AUTO_NEG
> > > speed: 1000 Mbps now, 1000 Mbps max
> > > LOCAL(ofs0): addr:00:0a:85:07:22:e9
> > > config: 0
> > > state: 0
> > > speed: 0 Mbps now, 0 Mbps max
> > >
> > >
> > >
> > > I senï½ packets from eth1 â LOCAL â eth2 and I hopeï½ that I can get nearly 1Gbps throughput.
> > >
> > > But I got only 100Mbps throughput.
> > >
> > > When I sent packets from eth1 â eth3 âeth2, I got nearly 1GHz throughput.
> > >
> > >
> > >
> > > Why I cannot get nearly 1GHz throughput when I use LOCAL port?
> >
> > I guess, when you say 1 GHz, you mean 1 Gbps.
> >
> > I don't understand what eth1 Ë LOCAL Ë eth2 means. The switch forwards
> > packets that arrive on one port to another port. So, it could for
> > example forward packets that arrive on eth1 to LOCAL. But I don't know
> > why packets sent to LOCAL would then go to eth2. It might be possible
> > to arrange for that with routing, but you didn't mention any routes.
> >
> > Maybe you should tell us what you're doing.
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20161130/27164348/attachment.html>
More information about the discuss
mailing list