Hi,
In general, if the number of loss history is less than 8, then it
would be more appropriate to consider the only available history when
you calculate the loss event rate.
But, even if we use 8 loss intervals to get the loss event rate having
only two of them are meaningful, it should not harm much for the
overall performance, which means it should not do badly in the steady
state environment.
If you only use the available (meaningful) histories to get the loss
event rate, then the protocol's adapting time to the steady state
would be decreased in time. Other than that, the performance of the
protocol is expected to remain same.
Soo-Hyun
P.S.
If you have a very high speed link and a very large bottleneck queue
with only small number of low speed application traffic, then you
probably want to consider the available history information when you
calculate the loss event rate.
On 4/20/06, Jeroen Van Velthoven <jeroen.vanvelthoven@xxxxxxxx> wrote:
Dear all,
I am performing some experiments with DCCP/TFRC. Therefore, I'm using the
DCCP implementation of Mattsson. My question relates to the calculation of
the average loss interval as it is done by this implementation. As
specified the TFRC RFC, the 8 most recent loss intervals are used to obtain
the average loss interval. However, also when the first loss event occurs
and in every situation where the number of initialized loss intervals is
strictly smaller than 8, this implementation takes the sum W_{tot} of all 8
weights and uses this result to determine the average loss interval.
I would assume that when the loss history contains only two loss intervals,
it is sufficient to take only these weights into account when calculating
W_{tot}. This would not result in an overestimate of the loss event rate
you get by taking into account the total of the 8 weights.
Can anyone confirm that the number of weights you need to use to obtain the
average loss interval equals the minimum of 8 and the number of the actually
created loss intervals?
Best Regards,
Jeroen
-------------------------------------------------------------------
Jeroen Van Velthoven
Phone: +32 (0)3 265.38.53
Fax: +32 (0)3 265.37.77
Email: Jeroen.VanVelthoven@xxxxxxxx
University of Antwerp (UA)
Dept. Mathematics and Computer Science
PATS Research Group
Campus Middelheim, building G
Middelheimlaan 1
B-2020 Antwerp
Belgium
-------------------------------------------------------------------
