Merit Network
Can't find what you're looking for? Search the Mail Archives.
  About Merit   Services   Network   Resources & Support   Network Research   News   Events   Home

Discussion Communities: Merit Network Email List Archives

North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: T1 Circuit actual throughput 1290Kbps

  • From: Sean Butler
  • Date: Thu Jul 09 08:16:37 1998

Tony Hariman wrote:
>Does anybody have experience having a T1 circuit with PPP
>encapsulation getting only 1290 Kbps maximum throughput looking
>at "sh int" result from cisco router or MRTG ?


Is it a long delay link running TCP traffic?  Long delay links
will always show significanlty lower utilization than the max
theoretical...

You can check out the following paper for more details, but
I'll give a quick example below.

   Lakshman&Madhow "The Performance of TCP/IP for Networks with
   High BWxDelay products and random loss" ,IEEE Trans. on
   Networking Vol5(3), june 97, pp329-335

   http://tesla.csl.uiuc.edu/~madhow/publications/ton97.ps


>From that paper, you can create the following chart:

Buffer Size     Link Utilization
60K             81%
120K            87%
180K            91%
480K            99%

The percentage utilization is based on the bandwidth * the delay,
so if you have a T1 at 1536K with a delay of 400ms, you get just over
600K  (very pessimistic... -- you can play with the numbers
based on the delay of the link.  My numbers are based on an analysis
I did of a satellite link some time ago...)

The buffer sizes are percentages of that 600K, at 10%, 20%, etc.

The buffer size is also the queue length * the max packet size
of the buffer....  So a Cisco with a default queue length of 40
and an MTU of 1500 = 60K, at which point you'd only see 81%
utilization for 1 TCP session...  That is just 1 session, so you'd
expect to see higher utilization for more sessions.

This all assumes FIFO queing...

Another point to consider is that with Internet traffic, there
are many many 'small flows' where it takes time to ramp up the
send window size, and before the time it would ramps up to use
the whole pipe, the flow is done

Sean Butler





Discussion Communities


About Merit | Services | Network | Resources & Support | Network Research
News | Events | Contact | Site Map | Merit Network Home


Merit Network, Inc.