CS848 Paper Review Form - Fall 2006

Paper Title: Database Replication for Dynamic Content Applications

Author(s): Gokul Soundararajan, Cristiana Azma and Ashvin Goel

1) Is the paper technically correct?
 [X] Yes
 [ ] Mostly (minor flaws, but mostly solid)
 [ ] No

2) Originality
 [ ] Very good (very novel, trailblazing work)
 [X] Good 
 [ ] Marginal (very incremental)
 [ ] Poor (little or nothing that is new)

3) Technical Depth
 [ ] Very good (comparable to best conference papers)
 [X] Good (comparable to typical conference papers)
 [ ] Marginal depth
 [ ] Little or no depth

4) Impact/Significance
 [ ] Very significant
 [X] Significant
 [ ] Marginal significance.
 [ ] Little or no significance.

5) Presentation
 [ ] Very well written
 [X] Generally well written
 [ ] Readable
 [ ] Needs considerable work
 [ ] Unacceptably bad

6) Overall Rating
 [ ] Strong accept (very high quality)
 [X] Accept (high quality - would argue for acceptance)
 [ ] Weak Accept (marginal, willing to accept but wouldn't argue for it)
 [ ] Weak Reject (marginal, probably reject)
 [ ] Reject (would argue for rejection)

7) Summary of the paper's main contribution and rationale
   for your recommendation. (1-2 paragraphs)

This paper describes a system for scheduling queries to a replicated 
database cluster.  The clusters are broken into sets depending on the
 application load and the SLA for that application.  They have a 
primary set which represents a disjoint set where reads occur and 
they have an overlapping set that executes write queries.  This 
method allows multiple machines to keep current data but to focus the
 workload on a smaller subset of those up-to-date replicas.

They make special considerations for the measuring of the latency and
 the adjustment of the set size based on dynamic loads.  First, they 
recognize there will be overhead when a new replica is brought into 
the set.  This will affect latency detrimentally for a short period 
of time.  During this time they do not measure latency.  They also 
stop oscillation by having a conservative two stage removal process.  
They verify their technique using the TPC-W and Rubis benchmarks.


8) List 1-3 strengths of the paper.  (1-2 sentences each,
identified as S1, S2, S3.)

S1 - They consider the affect of reallocation explicitly.
S2 - They stop isolation by conservatively removing replicas from 
the sets of replicas.
S3 - They accomplish partial overlap while maintaining disjoint sets 
for a majority of the workload.  This assumes that the workload is 
predominantly read queries.


9) List 1-3 weaknesses of the paper (1-2 sentences each,
identified as W1, W2, W3.)

W1 - The allocation formula is poorly described and not intuitive.
W2 - The description of the method of setting the threshold values 
was not concise and spread throughout the paper.



10) Detailed comments for authors.

I would have liked to see a little more detail in how the 
parameters of the system were defined.  Specifically, the 
threshold values, the alpha value and the time constant for 
replica setup.