[ns] Re: MRED algorithms under DiffServ - s parameter computation

[JORGE MANRIQUE MARGALEJO <jorge.manrique@upcnet.es>]

Hi Sergio,

I have not deeply studied the code you're talking about. However, as you explain in your mail, it seems reasonable to use 2 Mbps instead of 10 Mbps. Even if the "typical transmission time for a small packet" is strictly size/10Mbps, if you're limiting the bandwith to 2 Mbps you're somehow increasing the actual trasnmision time for that traffic and slowing down the "virtual link" for that traffic.

Using 2 Mbps sounds more logical to me but, as I say, I have not deeply studied the code so this is just my opinion.

Regards,

Jorge.

-------------------------


Hi! 
MRED algorithms are based on RED. In this, to compute average queue lenght 
when a packet enters an empty queue, the s parameter have to be computed (s 
in the Floyd paper, 1/PTC in ns implementation - DiffServ module). 
m is "the number of packets that might have been transmitted by the gateway 
during the time that the line was free" and s is "a typical transmission 
time for a small packet". 
My question is: 
e.g. 
I have an outgoing link of 10 Mbit/s in a DiffServ enabled router. An AF PHB 
is configured to have 2 Mbit/s of bandwidth and a MRED mechanism is applied 
to it. 
How should I compute s? Should I consider 10 Mbit/s or 2 Mbit/s? (I guess 2 
Mbit/s, but in ns 10 Mbit/s is considered). 
Thanks 
Sergio