Re: Asymmetry or others?


Hi,

#I have two linux (2.4) boxes in two universities connecting 
#via an Education WAN link. I got different bandwidth output
#if I put client and server on different boxes. One direction
#is about 15Mbps, and the other is 30Mbps. I use simple iperf
#(1.6.3) options:
#  a>$./iperf -s -w 100k
#  b>$./iperf -c a -w 100k

Without specifics about the end points in question, anything
would be pretty speculative, but hypothetically assume that
the hosts are connected via a moderately loaded DS3. A 15Mbps
differential in terms of available capacity (in vs. out) on 
that sort of link would not be inconceivable, and would account 
for the variation you're seeing. Do you have MRTG or other 
SNMP-based traffic measurements available for likely network 
choke points?

#Because it's education network, I really don't think it's
#asymmtric although I am asking the administrator about this.

We routinely see asymetric routes for higher ed sites, and
the behavior is only becoming more common as HPC connectivity, 
peering projects, statewide networks and carrier failures all 
increase the level of multihoming (and thus the variety of
possible routes) out there (season that stew with varying 
traffic management objectives including cost minimization, 
link utilization management, routing policy simplicity, 
politics, etc.). A trivial example of a gross asymmetry is
the traffic path between UO and UW (and vice versa):

traceroute to www.washington.edu (140.142.15.233), 30 hops max, 40 byte packets
 1  ge-4-2.uonet2-gw.uoregon.edu (128.223.142.3)  0.424 ms  0.403 ms  0.368 ms
 2  ge-0-0-0.0.uonet8-gw.uoregon.edu (128.223.2.8)  4.489 ms  7.126 ms  0.459 ms
 3  ge-0-0.core1.eug.oregon-gigapop.net (198.32.163.149)  0.423 ms  0.319 ms  0.471 ms
 4  eug-snva.oregon-gigapop.net (198.32.163.10)  13.581 ms  14.186 ms  12.977 ms
 5  snva-snvang.abilene.ucaid.edu (198.32.11.122)  18.053 ms  13.283 ms  13.009 ms
 6  sttl-snva.abilene.ucaid.edu (198.32.8.9)  33.913 ms  32.017 ms  35.897 ms
 7  hnsp1-wes-so-5-0-0-0.pnw-gigapop.net (198.48.91.77)  33.362 ms  31.435 ms  33.168 ms
 8  uwbr1-GE3-0.cac.washington.edu (198.107.151.51)  19.835 ms  18.957 ms  19.217 ms
 9  dirtdevil-V24.cac.washington.edu (140.142.154.15)  19.883 ms  20.595 ms  20.160 ms
10  www4.cac.washington.edu (140.142.15.233)  19.249 ms  19.152 ms  20.975 ms

vs. UW and UO:

1 dirtdevil-V18.cac.washington.edu (140.142.15.225) 0 ms 0 ms 1 ms 
2 uwbr2-GE2-1.cac.washington.edu (140.142.154.24) 1 ms 0 ms 0 ms 
3 core1-wes-ge-0-0-0-0.pnw-gigapop.net (198.107.150.119) 1 ms 1 ms 0 ms 
4 core1-pdx-so-0-0-0-0.pnw-gigapop.net (198.107.144.18) 4 ms 4 ms 4 ms 
5 prs1-pdx-FE2-0.pnw.gigapop.net (198.107.144.78) 4 ms 4 ms 5 ms 
6 198.107.144.90 (198.107.144.90) 4 ms 5 ms 4 ms 
7 ptck-core2-gw.nero.net (207.98.64.138) 5 ms 5 ms 5 ms 
8 eugn-core2-gw.nero.net (207.98.64.1) 7 ms 6 ms 7 ms 
9 eugn-car1-gw.nero.net (207.98.64.165) 7 ms 7 ms 7 ms 
10 207.98.64.34 (207.98.64.34) 21 ms 26 ms 21 ms 
11 ge-1-1.uonet2-gw.uoregon.edu (128.223.2.2) 20 ms 23 ms 20 ms 
12 darkwing.uoregon.edu (128.223.142.13) 21 ms 21 ms 23 ms 

A variety of traceroute gateways are available at http://www.traceroute.org/
if you're interested in testing additional pairs. 

So, I guess I would NOT assume that routes are generally symmetric. Hank 
Nussbacher has an excellent paper at 

http://www.internet-2.org.il/i2-asymmetry/index.htm

that you might to see for more on this.

Regards,

Joe



Other Mailing lists | Author Index | Date Index | Subject Index | Thread Index