[MLB-WIRELESS] Speed degradation over multiple hops

[Rob Clark]

Mark,

Yes but....

TCP/IP has built-in support for line delay..called windowing. Basically
- TCP can send packets without waiting for previous packet
acknowledgment. Most TCP stacks support one satellite delay
(600ms)....which is way longer than you would expect for wifi packet
delays.

However....it would be worth checking the size of the TCP window at each
end ... Maybe it is too low at one box. The window needs to be at
greater than xx bytes, where:

XX =  (link speed, bytes/sec) x (typical ping time, sec)

I am ASSUMING the traffic is being transferred via TCP (eg ftp).

Rob Clark


-----Original Message-----
From: owner-melbwireless at wireless.org.au
[mailto:owner-melbwireless at wireless.org.au] On Behalf Of Mark Aitken
Sent: Sunday, March 28, 2004 2:25 PM
To: Ben Short; melbwireless at wireless.org.au
Cc: mesh at itee.uq.edu.au; syd-wireless at lists.sydneywireless.com
Subject: Re: [MLB-WIRELESS] Speed degradation over multiple hops


At 03:19 PM 3/28/04 +1000, Ben Short wrote:


>Now, each link from hop to hop you can send or recieve at about 
>400kb/s,
>which is to be expected. However if I download
>direction from hop 6 to hop 1, I am usually lucky to get 10kb/s. There
is 
>no other traffic on the network at the time, and
>hidden nodes are not an issue. I am assuming interference is not an
issue, 
>because each link can do 400kb/s.


It only stands to reason that each frame has to be totally received
before 
it can be re-transmitted at each hop. Therefore there is a finite amount
of 
time that is required for each system to receive and re-transmit the 
frame.  This can drastically affect the through put, especially if for
what 
ever reason a frame is lost in transit and needs to timeout before it is

re-transmitted.

Just my 2c worth.

Mark



To unsubscribe: send mail to majordomo at wireless.org.au
with "unsubscribe melbwireless" in the body of the message