|
Some thoughts on improving the Faster Restart draft
(http://www.ietf.org/internet-drafts/draft-ietf-dccp-tfrc-faster-restart-02.txt)
: 1) I have two thoughts about using the receiver rate
adjustment algorithms: ·
I believe that we may not need the
receiver rate adjustment algorithms for TFRC after idle periods as the receiver
rate is calculated only for 1 RTT and the deflated rate may not be that
significant, whereas we “may” have the receiver rate adjustment
algorithms for CCID 3 since the receiver rate is reported over the time since
the last feedback report. I am not sure why we have these two different
behaviors for TFRC and CCID 3!
2) X_active_min_rate := min(8*s, max(4*s, 8760 bytes)). Should be X_active_min_rate := min(8*s, max(4*s, 8760 bytes))/RTT? 2) Maintaining the X_active_min_rate during data-limited
period: /* First phase. Adjust X_recv so send rate doesn't drop below X_active_min_rate as the result of an idle and/or slow period. */ If the feedback packet does not indicate a loss or mark and the old X_recv >= X_active_min_rate/2, then X_recv := max(X_recv, X_active_min_rate/2).
I did these modifications in ns-2 and did some tests to find
out that this is too conservative in the presence of both packet loss and
data-limited period and the performance is worse than TFRC (RFC3448-bis). RFC3448-bis-02c states If (feedback packet reports Limited Receive Rate or sender has been data-limited over period covered by the last feedback packet) min_rate = max(2*X_recv, W_init/R); Else // typical behavior min_rate = 2*X_recv;
So you can see that even though in the presence of packet
loss and during a data-limited period, the min-rate for TFRC would be atleast a
min of W_init/R. This is not the case with FR. As maintaining the min_rate or X_active_min_rate would be
limited by the throughput equation. I would suggest removing the loss or mark
criteria while considering the X_active_min_rate for idle or during
data-limited period. /* First phase. Adjust X_recv so send rate doesn't drop below X_active_min_rate as the result of an idle and/or slow period. */ If the old X_recv >= X_active_min_rate/2, then X_recv := max(X_recv, X_active_min_rate/2).
So I basically suggest that even in the presence of loss, we
could maintain the X_recv to atleast the X_active_min_rate as the sending rate
would be then limited to the throughput equation. So it shouldn’t be a
problem. Any suggestions? -Arjuna ------------------------------- Dr.Arjuna Sathiaseelan Electronics Research Group Email: arjuna@xxxxxxxxxxxxxx Web: www.erg.abdn.ac.uk/users/arjuna Phone : +44-1224-272780 |
