nanog mailing list archives

Re: Provider Bypass


From: "Christopher Wolff" <chris () bblabs com>
Date: Sun, 30 Sep 2001 19:06:33 -0700


Joe;

Thank you for the additional information...this application and information will be quite helpful.  

I'll run these tests and apply some of the guidance in the network tuning page you forwarded and let you know what 
happens.

Regards,
Christopher
---------- Original Message ----------------------------------
From: Joe St Sauver <JOE () OREGON UOREGON EDU>
Date: Sun, 30 Sep 2001 17:38:05 -0700 (PDT)

two paths one through Phoenix and on=
e through Houston, and one path can sustain 250KB ftp transfers but t=
he other only sustains 50kB ftp transfers...

Based on trips on I10, my recollection is that there's a substantial
distance between the two locations; assuming that distance translates
to increased latency, I wouldn't be surprised to see a difference in
throughput.

I would suggest tuning the hosts to handle high bandwidth flows per
http://www.psc.edu/networking/perf_tune.html and then running a known-
capable throughput testing program such as iperf 
( http://dast.nlanr.net/Projects/Iperf/ )

I would think there is a way for us to static route this traffic thro=
ugh Phoenix, or hardcode it in the BGP table...I tried a few tweaks w=
ith no joy.

I'm assuming you're multihomed to both Phoenix and Houston? Is the issue 
inbound or outbound? Depending on the direction, you should be able to 
control the traffic using prefix prepending or MEDS, assuming your provider 
is willing to play. Sam Halabi's routing book from Cisco Press has nice 
examples?

Or are you trying to do traffic engineering over a single connection,
within your provider's backbone? (If that, I think you're headed for
frustration -- I'd suggest a different NSP :-))

Regards,

Joe



Current thread: