[net-perf] Performance test results

Michael Van Norman mvn at ucla.edu
Mon Jun 13 16:33:35 PDT 2005


I did some performance tuning on my side last Friday night.

Turning off tcp segmentation offload was a big winner.

/Mike

Mike Iglesias wrote:
>>One observation is that at 1500-byte MTU, saturation level appears to
>>be 941 mb/s of data on a gig link and 94.1 mb/s of data on a fastE
>>link. In other words, when iperf is reporting 941 mb/s of data
>>throughput (not counting protocol overhead), the router/switch
>>interface is reporting 1000 mb/s utilization.  (Strangely, in the
>>iperf 2.02, I am also seeing up to 943 mb/s on the same link, so
>>perhaps it's doing slightly different math in the later version.)
> 
> 
> I've seen ~940mb/s between two systems on the same card/vlan on a
> Cisco 6500 router.
> 
> I ran a test between UCI and UCLA with the UCI system's mtu set to 8192
> (I need to see what the real maximum in Linux is, that worked for now).
> I got
> 
> UCI -> UCLA   849mb/s
> UCLA -> UCI   719mb/s
> 
> using "-2m -i 2 -t 10".
> 
> What I'm wondering is why changing that would make such a big difference?
> So I went and tried it from another system inside campus that still has
> it's mtu set to 1500 (we can't change that one due to limitations on the
> Cisco card it is plugged into right now).  Here's what I got:
> 
> UCI -> UCLA   214mb/s
> UCLA -> UCI   630mb/s
> 
> Those number are way higher than what I was seeing late last week, so
> what has changed between UCI and UCLA?  We're talking about 10x faster
> from UCI to UCLA and 2-3x faster from UCLA to UCI.
> 
> 
> Mike Iglesias                          Email:       iglesias at draco.acs.uci.edu
> University of California, Irvine       phone:       949-824-6926
> Network & Academic Computing Services  FAX:         949-824-2069
> 
> 
> test systems:  iperf-uci.nac.uci.edu and bwctl-1.border.ucla.net
> 
> UCI -> UCLA
> 
> RECEIVER START
> 3327688445.177039: /usr/local/bin/iperf -B 169.232.34.34 -P 1 -s -f b -m -p 5004 -w 2097152 -t 10 -i 2
> ------------------------------------------------------------
> Server listening on TCP port 5004
> Binding to local address 169.232.34.34
> TCP window size: 4194304 Byte (WARNING: requested 2097152 Byte)
> ------------------------------------------------------------
> [ 14] local 169.232.34.34 port 5004 connected with 128.195.120.5 port 5004
> [ 14]  0.0- 2.0 sec  211924900 Bytes  847699600 bits/sec
> [ 14]  2.0- 4.0 sec  212030720 Bytes  848122880 bits/sec
> [ 14]  4.0- 6.0 sec  210280620 Bytes  841122480 bits/sec
> [ 14]  6.0- 8.0 sec  213756400 Bytes  855025600 bits/sec
> [ 14]  8.0-10.0 sec  214334340 Bytes  857337360 bits/sec
> [ 14]  0.0-10.0 sec  1064869888 Bytes  849811323 bits/sec
> [ 14] MSS size 8140 bytes (MTU 8180 bytes, unknown interface)
> RECEIVER END
> 
> 
> UCLA -> UCI
> 
> RECEIVER START
> 3327688507.491709: /usr/local/bin/iperf -B 128.195.120.5 -P 1 -s -f b -m -p 5002 -w 2097152 -t 10 -i 2
> ------------------------------------------------------------
> Server listening on TCP port 5002
> Binding to local address 128.195.120.5
> TCP window size: 4194304 Byte (WARNING: requested 2097152 Byte)
> ------------------------------------------------------------
> [ 14] local 128.195.120.5 port 5002 connected with 169.232.34.34 port 5002
> [ 14]  0.0- 2.0 sec  185390164 Bytes  741560656 bits/sec
> [ 14]  2.0- 4.0 sec  185022200 Bytes  740088800 bits/sec
> [ 14]  4.0- 6.0 sec  182360420 Bytes  729441680 bits/sec
> [ 14]  6.0- 8.0 sec  172226120 Bytes  688904480 bits/sec
> [ 14]  8.0-10.0 sec  174440200 Bytes  697760800 bits/sec
> [ 14]  0.0-10.0 sec  902602752 Bytes  719585241 bits/sec
> [ 14] MSS size 8140 bytes (MTU 8180 bytes, unknown interface)
> RECEIVER END
> 
> 
> ----------------------------------------------------------------------------
> test systems:  iperf-calit2.nac.uci.edu and bwctl-1.border.ucla.net
> 
> UCI -> UCLA
> 
> RECEIVER START
> 3327690535.113290: /usr/local/bin/iperf -B 169.232.34.34 -P 1 -s -f b -m -p 5002 -w 2097152 -t 10 -i 2
> ------------------------------------------------------------
> Server listening on TCP port 5002
> Binding to local address 169.232.34.34
> TCP window size: 4194304 Byte (WARNING: requested 2097152 Byte)
> ------------------------------------------------------------
> [ 14] local 169.232.34.34 port 5002 connected with 128.195.158.11 port 32836
> [ 14]  0.0- 2.0 sec  52717248 Bytes  210868992 bits/sec
> [ 14]  2.0- 4.0 sec  53792496 Bytes  215169984 bits/sec
> [ 14]  4.0- 6.0 sec  53769240 Bytes  215076960 bits/sec
> [ 14]  6.0- 8.0 sec  53791128 Bytes  215164512 bits/sec
> [ 14]  8.0-10.0 sec  53791128 Bytes  215164512 bits/sec
> [ 14]  0.0-10.1 sec  270532608 Bytes  214299549 bits/sec
> [ 14] MSS size 1368 bytes (MTU 1408 bytes, unknown interface)
> RECEIVER END
> 
> 
> UCLA -> UCI
> 
> RECEIVER START
> 3327690483.405262: /usr/local/bin/iperf -B 128.195.158.11 -P 1 -s -f b -m -p 5008 -w 2097152 -t 10 -i 2
> ------------------------------------------------------------
> Server listening on TCP port 5008
> Binding to local address 128.195.158.11
> TCP window size: 4194304 Byte (WARNING: requested 2097152 Byte)
> ------------------------------------------------------------
> [ 14] local 128.195.158.11 port 5008 connected with 169.232.34.34 port 5008
> [ ID] Interval       Transfer     Bandwidth
> [ 14]  0.0- 2.0 sec  107480888 Bytes  429923552 bits/sec
> [ 14]  2.0- 4.0 sec  173785088 Bytes  695140352 bits/sec
> [ 14]  4.0- 6.0 sec  166240696 Bytes  664962784 bits/sec
> [ 14]  6.0- 8.0 sec  170231512 Bytes  680926048 bits/sec
> [ 14]  8.0-10.0 sec  170154496 Bytes  680617984 bits/sec
> [ 14]  0.0-10.0 sec  789815296 Bytes  630442316 bits/sec
> [ 14] MSS size 1368 bytes (MTU 1408 bytes, unknown interface)
> [ 14] Read lengths occurring in more than 5% of reads:
> [ 14]  1368 bytes read 203891 times (72.7%)
> [ 14]  8192 bytes read 52393 times (18.7%)
> RECEIVER END
> 
> 
> _______________________________________________
> Network-performance mailing list
> Network-performance at lists.cenic.org
> http://lists.cenic.org/mailman/listinfo/network-performance


More information about the Network-performance mailing list