[Planetlab-users] A lot of nodes completely overloaded?

[Jeroen Vlek]

L.S.,

I'm trying to run a routing experiment in my slice. However, most nodes
are completely overloaded. Here a random sample of nodes from my slice
zib_proximity:

planetlab-2.man.poznan.pl 20.22 18.75 18.70 2/831 15405
planetlab1.warsaw.rd.tp.pl 2.52 2.45 2.55 12/701 14928
planetlab2.sics.se 5.71 7.60 7.00 2/942 14766
planetlab-1.ssvl.kth.se 3.87 4.38 5.75 2/1087 28253
planetlab3.mini.pw.edu.pl 10.65 9.70 8.99 7/840 32238
host1.planetlab.informatik.tu-darmstadt.de 12.15 14.30 15.20 7/933 26369
planetlabtwo.ccs.neu.edu 1.70 1.61 1.59 4/1667 9932
planetlab1.cs.stevens-tech.edu 1.75 2.55 2.61 8/1246 29168
75-130-96-13.static.oxfr.ma.charter.com 1.74 3.15 3.31 6/1153 6427
planetlab01.erin.utoronto.ca 2.14 2.28 2.33 2/899 22419
righthand.eecs.harvard.edu 21.99 17.24 15.59 20/2411 2238
planetlab2.cs.dartmouth.edu 17.56 17.19 17.22 2/1489 14476
kupl2.ittc.ku.edu 12.30 17.16 16.84 12/1822 13499
planetlab-1.vuse.vanderbilt.edu 13.46 13.32 12.18 2/977 29653
server2.planetlab.iit-tech.net 10.34 10.86 11.23 2/919 8097
planetlab3.ucsd.edu 4.77 4.92 6.04 7/960 19901
planetlab2.flux.utah.eduplanetlab2.cs.ucla.edu 25.96 28.88 31.00 6/1199 21021
planetlab1.postel.org 14.12 13.64 14.61 19/1212 31929
pl2.unm.edu 3.62 3.59 3.62 18/1068 15741
 8.79 11.18 13.40 3/1763 9761
planetlab2.cs.duke.edu 17.66 21.50 21.64 6/1265 3022
planet-lab2.ufabc.edu.br 1.02 1.05 1.07 3/392 21840
recall.snu.ac.kr 10.14 11.05 11.64 6/839 20244
planetlab-01.naist.jp 13.54 12.31 12.14 12/1254 15888
planetlab3.singaren.net.sg 1.74 2.42 3.42 8/1116 32029
planetlab1.cosc.canterbury.ac.nz 1.78 2.25 1.78 5/778 31817
planetlab2.eecs.umich.eduplanetlab2.cis.upenn.edu 17.72 18.03 14.46
23/2460 29086
 30.29 30.05 32.30 46/6499 2426
kc-sce-plab1.umkc.edu 4.75 6.85 7.08 2/1189 9706
ds-pl3.technion.ac.il 2.34 2.36 2.20 5/1026 25957
planet1.cs.huji.ac.il 18.16 21.96 24.44 14/1058 29632
planetlab2.ci.pwr.wroc.pl 16.96 17.69 21.48 7/934 4748
planetlab1.iitr.ernet.in 0.48 0.65 0.71 2/958 24987
plnode02.cs.mu.oz.au 0.49 0.35 0.32 7/1005 28614
planetlab1.eecs.jacobs-university.de 33.68 34.36 34.83 3/1926 19677


The point is that it takes longer to calculate a response then it does to
route the lookup. That not only influences the measurements, our Chord#
bootserver also believes a lot of nodes to be dead, because timeouts
constantly occur.

Does anyone have an idea what causes these big (and unfair!) loads? And
what can be done about it?


Jeroen Vlek