[lug] Traceroute question

[Michael Deck]

At 04:35 PM 12/13/00 -0700, D. Stimits wrote:
>Deva Samartha wrote:
> >
> > But the 200 ms times are not right for a 114kb DSL, that should be in the
> > 50's, unless he has a 1:5 ratio out to in.
> >
> > S
>
>DSL gives large throughput, and traceroute says nothing about
>throughput. It only tests with 3 tiny ICMP packets. 114kb is an average,
>and does not include the time it takes to fill the pipe...the latency.
>If I was measuring throughput in terms of cars driving from city to city
>(borrowing from other posts), I could measure how fast a car gets there.
>Or I could measure effectiveness by counting the number of passengers
>transported over time. A porsche 930 turbo could get 3 passengers from A
>to B quickly, and win one performance test...latency. A bus with 75
>passengers would win the effectiveness...throughput. DSL 114kb is only a
>measure of throughput; it downloads large amounts of data quickly, but a
>small data packet on DSL might suck compared to a small packet on ISDN,
>or even a 56k modem.

True. The real problem is DNS lookups. Apps like nslookup and Netscape seem 
to be timing out waiting for name resolution and I'm thinking that this 
might be contributing to the problem. If it turns out that ISDN was giving 
me poor throughput on large transfers but great DNS and small-packet rates, 
then I'm probably going back to ISDN because most of what I seem to be 
doing is lots and lots of small packets (e.g. every little bitmap on the 
slashdot main page) and the startup effects are gonna kill me not to 
mention the DNS failures.

-M

Michael Deck
Cleanroom Software Engineering, Inc.