On Wed, 14 Jan 1998, kim yong sin wrote:
> I'm trying to simulate satellite links using ns1.3.
> Before 'TCP extension option' is included, I've just changed parameters
> (delay, bandwidth).
> Now, I hesitate simulation.
>
> 1. What is the value of 'window size' ?
> Using TCP window scaling option, I can extend window size to 2^30 bytes.
> But, window scaling option is included in TCP SYN segment.
> So, TCP has no information about window scaling.
> Can I use bandwidth*delay as window size?
>
> 2. What is the value of 'initial ssthresh' ?
> Before TCP extension option is included, initial ssthresh equals to
> 'window size' from receiver.
> Still can I use 'window size' as initial ssthresh?
>
> 3. As you know, most people assume that they don't use delayed ACK in their
> reports.
> If they assume like that, these reports can explain real TCP congestion
> control in satellite links?
>
>
> sample topology)
>
> hosts gateway terminal
> O-----------------------O----------------------O
> 10 Mbps, 10ms (satellite link)
> up-downlink : 10 Mbps, 290 ms (RTT = 580 ms)
> * MSS : 1000 bytes (or other).
> * use window scaling option.
>
> 4. Above figure shows a sample topology.
> What are the parameter values (i.e window size, initial ssthresh, etc)
> in your simulation or experiment.
>
I suggest:
1. You must add error link module in the satellite link between gateway
and terminal, and the burst error pattern is a more sophisticated
than random error pattern.
2. If you don't use window scale option, the utilization is too low for
10 Mbps in your simulation.
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:35 EST