Re: Re: [OT] WAN throughput comparisons

["Dean Smith" <dean@xxxxxxxxxxxxxxx>]

Exactly.

As bandwidths increase, latency becomes a bigger issue for single session
Data transfers.

For example we had a situation where a customer was complaining they
couldn't use their 34 Mb/s serial circuit fully. On digging we established
the end-to-end latency between test hosts was 14ms. For a single TCP
session between windows hosts using default TCP/IP settings this gives a
max throughput of about 8 Mbits/sec. (from memory)

This is due to the tcp window size - the sending host will send only so
much data before waiting the ack to arrive back from the recieving host. In
order to send more whilst waiting for the ack, the windows size need
increasing. They tweaked the necessary registry settings and magically
increased the WAN throughput.....

Dean
----- Original Message -----
From: David Buckley
To: ukha_d@xxxxxxx
Sent: Tuesday, July 27, 2004 1:35 PM
Subject: [ukha_d] Re: [OT] WAN throughput comparisons


--- In ukha_d@xxxxxxx, "Paul Gordon" <paul_gordon@h...>
wrote:
> Anyone know of a reference guide to the "typical" data
> throughput figures on various kinds of WAN links?

What sort of stuff you interested in?  I used to do a lot of this
sort of analysis.  There is a lot of "it depends" stuff in there
:-)

Some variables in order of importance:

Latency of link
Software performing the data transfers (eg ftp vs telnet)
Bandwidth
Number of parallel network transactions

The raw throughput in bytes is roughly the bits per second of the
line divide by ten, but the big questions are can that bandwidth be
effectively used.