Reviewer: 1


   Originality : 4
   Relevance : 2
   Presentation : 4
   Recommendation : 4

Summary: This paper is an application of passive measurement
for Qos scheduling.  The authors present a framework that
combines small Flow Completion Time of short-lived flows and
fair bandwidth allocation of long-lived flows using a
resource allocation algorithm Markov Active Yield (MAY).
The paper presents a theoretical analysis and simulation
study to validate their results. The paper is nicely written
and well presented.

Details:


Candidate for the best paper award? : No

=====================================

Reviewer: 2


   Originality : 3
   Relevance : 3
   Presentation : 4
   Recommendation : 2

Summary: Propose keeping track of drop history of flows to
regulate traffic at a switch. Analyze the proposed scheme to
show that fair sharing is feasible.

Details: Drop history has been used as a guide for punishing
non-responding flows earlier in RED-PD[13] and other work
not cited here.

Probabilistic state maintenance or maintaing partial state
only for large flows is proposed earlier in SRED, LRU-FQ and
SACRED. Authors do not seem to be aware of this work.

Authors need to compute the complexity of the proposed
approach. Every Delta units, all the state is being scanned
to update the various state variables being maintained, as
shown in Fig. 1.

The strength of the paper is in the analysis portion of the
paper.


Candidate for the best paper award? : No

=====================================

Reviewer: 3


   Originality : 4
   Relevance : 3
   Presentation : 4
   Recommendation : 5

Summary: This paper proposes MAY - the algorithm for
resource allocation that uses the knowledge of past drops
per flow to estimate future drop probability. MAY keeps
small amount of state but achieves good fairness, as shown
through extensive simulations. I found that this paper had
the right blend of theory and simulations, it discusses an
important problem and proposes a simple, elegant solution.

Details: Page 1: small FCT short-lived -> small FCT of
short-lived Page 2: To end this -> To this end

In section 2-A you gave the estimate of how many flows could
be stored in the hash table but we still don't know if this
is sufficiently large to  support number of flows w drops at
Interent scale. Such data would much enhance your claim of
implementation feasibility.

Is Figure 3 obtained in ns2 simulations? If so we'd need to
know more details about the simulation environment: did you
use some traces, which traces, how did you use them in ns2,
etc. If the Figure is obtained using a customized simulator
of the simple model of TCP with delayed acks we need more
details about this too.

Page 5: payed -> paid

RED line is nearly invisible in Fig 4. Same goes for max-min
dotted line in Fig 6.

Page 7: We varied -> We vary
        or
        and evaluate -> and evaluated
For Fig 7 it would be nice to see the throughput of TCP
flows also, not just  the CBR flow

Page 8: typical RTT -> typical RTTs or typical RTT values
        payed -> paid


Candidate for the best paper award? : Yes