---------------------------------------------------------------------
CS 8803 AIC Paper Review Form
---------------------------------------------------------------------
Reviewer's Name and affiliation 

Wouldn't you like to know :)

---------------------------------------------------------------------
Paper Title:

Using Prediction to Accelerate Coherence Protocols

---------------------------------------------------------------------
Overall evaluation of work and contribution [1-10] : 5

Scale Definition:
10=Excellent work and a major contribution
 7=Good solid work of some importance
 5=Solid work but marginal contribution
 3=Marginal work and very minor contribution
 1=Very questionable work and contribution

----------------------------------------------------------------------
Originality / Novelty  [1-10] : 7

Scale Definition:
10=Trailblazing
 7=A pioneering piece of work
 5=One step ahead of the pack
 3=Yet another paper about ...
 1=Ancient Egyptians did this first

----------------------------------------------------------------------
Relevance to the class [1-10] : 10

Scale Definition:
10=Right on target
 8=Definitely relevant
 4=Close enough
 3=Not really appropriate for this class
 1=Definitely inappropriate for this class

---------------------------------------------------------------------
Writing quality, Readability and Organization
Your Rating [1-10] : 7

Scale Definition:
10=Very good
 7=Basically well written
 5=Readable
 3=Needs considerable work
 1=Completely incomprehensible

---------------------------------------------------------------------
Overall recommendation [1-10] : 5

Scale Definition:
10=Definitely accept (very high quality)
 7=Accept (good quality)
 5=Accept if room (marginal quality)
 3=Weak reject
 1=Accept only over my dead body

---------------------------------------------------------------------
Summary
[Please summarize the paper in your own words and very few sentences.]

This paper proposes Cosmos coherence message predictor to accelerate
coherence protocols. Cosmos uses two-level adaptive branch predictor to
record the previous history. By recording previous history of coherence
message exchange and executing the expected action speculatively, latency
can be hidden when next message is correctly predicted. Profiling study
shows that Cosmos' prediction accuracy high.

----------------------------------------------------------------------
What is the strength of the paper? (1-3 sentences):

The main contribution of Cosmos is that it can recognize a message
exchange pattern not known a priori, and adapt itself to predict next
coherence message. Memory overhead is well addressed, which must be one
of the main drawback of Cosmos.

---------------------------------------------------------------------
What is the weakness of the paper? (1-3 sentences):

Although the title says that Cosmos is aimed to accelerate coherence
protocols, there's no simulation or result on the execution time. There's
just a sensitivity study without actual delay estimates. Even though
prediction accuracy is 99%, overall performance might be worse if the
recovery from mis-speculation takes a long time. At least, it must be
shown that the time to recover from mis-speculation is not long.

---------------------------------------------------------------------
Questions you would like the authors to address before you make
a final decision whether to accept this paper to the conference:

How long will it take to recover from mis-speculation?

---------------------------------------------------------------------
Comments (will be returned to the author(s))
[Comment on strengths and weaknesses of the paper; in particular 
 note any ways in which the paper could be improved.]

It would be better if you show the impact of Cosmos on the execution time.

---------------------------------------------------------------------
Confidential Comments (will NOT be seen by the author(s))

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