Hi Srini, Thanks for your reply. I just tested this on the same HP just rewriting the DL_DST and the performance is the same there too so it must be done in software completely on these switches :(
I'll try this out on an NEC and pronto hopefully it will be faster there. Thanks, Aaron On Sun, Jun 26, 2011 at 1:45 PM, Srini Seetharaman <seeth...@stanford.edu> wrote: > I think this is expected. Most switches perform the NW_DST rewrite in > software, thereby having a low throughput. You should consult the > release notes of the switch you're using. > > With some Broadcom-based switches (e.g., NEC or Pronto) the DL_DST > rewrite is done is hardware. So, if you rewrite only that field, it > might work at line rate. > > NetFPGA switches perform NW_DST rewrite in hardware. I'm not sure > about the TCP_DST. > > On Sun, Jun 26, 2011 at 10:38 AM, Aaron Rosen <aro...@clemson.edu> wrote: >> Also, if I only modify DL_DST, NW_DST, (and not the TCP_DST port) it >> does not seem to improve performance. >> >> Thanks, >> >> Aaron >> >> On Sun, Jun 26, 2011 at 1:33 PM, Aaron Rosen <aro...@clemson.edu> wrote: >>> Hello, >>> >>> I'm using openflow to match a TCP packet and then modifying the packet >>> to get it to connect to a different host. >>> >>> To do this I need to modify the DL_DST, NW_DST, TP_DST (to get the >>> packets to the new host) and to get them back to the original host I >>> need to modify (DL_DST, NW_SRC, TP_SRC). Though I find when I do this >>> the connection is slowed down greatly. >>> >>> Here is the flow entry that is getting installed to modify the >>> connection to get it to the new host. >>> >>> cookie=0, duration_sec=13s, duration_nsec=623000000s, table_id=2, >>> priority=32768, n_packets=719, n_bytes=1085694, >>> idle_timeout=50,hard_timeout=0,tcp,in_port=36,dl_src=00:1b:21:43:a1:e4,nw_src=10.42.100.105,nw_dst=10.42.100.96,tp_src=48480,tp_dst=5001,actions=mod_dl_dst:00:1b:21:75:7a:92,mod_nw_dst:10.42.100.104,mod_tp_dst:9877,output:39 >>> >>> >>> When I use iperf to measure the bandwidth I find that if i use packet >>> rewrite I get 645Kbit/sec and If I don't I get 683 Mbits/sec. This >>> traffic is passing though an HP ProCurve J8693A Switch 3500yl-48G. >>> >>> >>> Is this what is expected? >>> >>> Thanks, >>> >>> Aaron >>> >>> -- >>> Aaron O. Rosen >>> Masters Student - Network Communication >>> 306B Fluor Daniel >>> >> >> >> >> -- >> Aaron O. Rosen >> Masters Student - Network Communication >> 306B Fluor Daniel >> _______________________________________________ >> openflow-discuss mailing list >> openflow-discuss@lists.stanford.edu >> https://mailman.stanford.edu/mailman/listinfo/openflow-discuss >> > -- Aaron O. Rosen Masters Student - Network Communication 306B Fluor Daniel _______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
