CS848 Paper Review Form - Fall 2006

Paper Title: Dynamic Provisioning of Multi-tier Internet Applications

Author(s):

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
 [X] Very well written
 [ ] 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 presents a dynamic provisioning mechanism for multi-tier internet 
applications based on three techniques i) predictive provisioning, that helps
provision resources ahead of time for long-term variations such as time-of-day
or seasonal effects ii) reactive provisioning, which helps predict and react 
to the short-term fluctuations in web workloads such as flash crowds iii) request 
policing, which helps to ensure SLAs by turning down additional requests in 
case all the available resources have been utilized already. Through experimental 
evaluation it has been shown that all the three mechanisms listed above are necessary 
for effective provisioning in multi-tier web applications.

The idea presented in this paper is unique because it models the servers at different 
tiers(can be arbitrary in number) as network of queues and then uses the G/G/1 queuing 
system theory to predict inter-arrival and service times for the requests; calculating 
the number of servers to be allocated at each tier for peak loads. The mechanism 
presented also conveniently handles session based workloads which are very common 
in today’s internet applications.


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

S1: Incorporated mechanisms to handle both long term (on the order of days/hours) workload 
fluctuations as well has short term peaks (on the order of minutes) and shows the 
effectiveness of using them together.

S2: Effectively uses the G/G/1 queuing system to define an analytical model of an n-tier 
internet application which is used to predict the number of servers to allocate at each 
tier for handling peak loads.

S3: The work is very well written with each section leading smoothly into the next. 
The paper also effectively used examples to make its point.

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

W1: The paper does not give sufficient details of the provisioning of the control plane 
itself. From the information provided, it seems to be a single point of failure and would 
be hard to scale for an n-tier web application with arbitrary number of servers reporting 
to the control plane.

W2: The error estimation of the architecture presented is based on noticeable underestimates 
of predictive provisioning only. It could have been useful by correcting the predictions in 
case of major overestimates so as to more efficiently utilize the available resources.

W3: The paper does not make clear that how the proposed architecture caters the allocation 
of varying resources i.e. not all the servers in the system have equal capabilities in terms 
of processing power/memory/disks etc.


10) Detailed comments for authors.