CS848 Paper Review Form - Fall 2006

Paper Title: Triage: Performance Differentiation for
Storage Systems Using Adaptive Control

Author(s): MAGNUS KARLSSON, CHRISTOS KARAMANOLIS, and XIAOYUN ZHU


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 proposes a way to ensure performance isolation and differentiation 
in data hosting center. Unlike solutions proposed in other literatures, the paper 
exhibits a solution for management requires no assumption about workload characteristics. 
The paper demonstrates the design of a on-line feedback loop with an adaptive controller 
that can periodically change its configuration (parameters) rather than a nonadaptive controller 
that is designed offline and once set up parameters never changed. Since it is very hard 
to model the database system behavior, those nonadaptive controllers cannot behave well. 
Also the controller is distributed to ensure high availability.

    This paper focus on throughput when it measure system performance. For each workload the paper 
tries to achieve two goals: latency and maximum throughput. The paper notices that within one workload 
the customer usually doesn¡¯t ¡°assign the same importance to the entire range of throughput they require 
for their workloads.¡± Therefore the paper designs a way to allocate different hosts with in different 
phases of one workload to optimize throughput. 


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

S1: The solution requires no provisioning on resources and makes no assumptions on system characteristics.

S2: The solution notices that customer may have different preference over the entire throughput session.

S3: The paper is well-written.


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

W1: The paper assumes there is somebody else solves the optimization problem of achieving high utility 
by differentiating workloads.


10) Detailed comments for authors.

    Just a little concern about using different hosts to share one workload. It is true for some applications 
that they assign different value among the entire throughput. Usually the first few IOs are very important 
for them to make progress. However, what if those IOs contains some updating transactions that write some 
data to the database, and later IOs rely on those updates. If it happened to be that the first IOs update 
some entries in one host and later IOs request data from another host, how can it guarantee the consistency 
of the data? I think the author assumes that the underlying database doesn¡¯t use lazy update among its 
replicas.