[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Canadian mirror ftp3.nrc.ca - throughput issues



Adam Majer un jour écrivit:
  132.246.2.23 - ftp3.nrc.ca

This mirror is included as part of ftp.ca.debian.org


The connection through telus seems very slow,

14  nrc-gi-1-7.nrnet2.nrc.ca (132.246.4.133)  55.036 ms * *
15  ftp3.nrc.ca (132.246.2.23)  56.391 ms nrc-gi-1-7.nrnet2.nrc.ca
(132.246.4.133)  57.157 ms ftp3.nrc.ca (132.246.2.23)  56.539 ms



While connection through allstream is full speed,


AFAIK, Allstream is very well connected with most (if not all) main canadians networks while the connectivity from Cogent with both Ca*NET and Videotron seems to be quite bad.

I made some quick testing with my mirrors and others places, and it seems that Congent either have some problems or have a much poorer connectivity with canadians networks than Allstream.

That said, there might be some other problems on NRC side as well. Here are some bandwidth tests I have made.


From my mirror to NRC (gulus.USherbrooke.ca)
 NRC   => 1500 K/s

 Could be faster, but it seems reasonnable.

From Videotron to NRC:
 gulus => 1050 K/s (about the bandwidth I have from Videotron)
 NRC   =>   40 K/s (????)

What is going on? My packets are going from Videotron to an internet node on CA*Net (the best route), then move to bell.ca to go back to NRC in 17 hops, while my mirror connect to it directly using CA*Net in 7 hops.

  Looks like to me that there is a routing problem somewhere.

  Also, the 40 K/s is probably due to traffic shapping somewhere.


Could you do another traceroute with my mirror? For what I can see, your connection will probably avoid using Cogentco if possible.


And also, could you do a wget of some big files to have an idea of the speed you get:

wget http://gulus.USherbrooke.ca/debian/dists/etch/Contents-i386.gz
wget http://ftp3.nrc.ca/debian/dists/etch/Contents-i386.gz


I won't be surprised if you also have a much slower connection to my mirror when using Telus instead of Allstream, but I expect that the number of hops will be much lower than the 15 you get from NRC. That said, I am curious of the results.


I don't know what is going on. Maybe it is cogent, but this has been the
problem for *months* now.


Another possibility is that NRC have little money for commercial Internet bandwidth, tought it won't explain the crazy route I sometime get from Videotron. Now, if you have about the same results with both mirrors, then it would at least means that the problem is not NRC specific.

Simon Valiquette
http://gulus.USherbrooke.ca


Reply to: