incubator-hama-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Edward J. Yoon" <edwardy...@apache.org>
Subject Our paper has been accepted as a workshop paper at IEEE CLOUDCOM'2010
Date Thu, 30 Sep 2010 10:18:59 GMT
You can download that paper here
https://blogs.apache.org/hama/entry/hama_in_academic_paper

And, If you have some feedback about this, just reply here or directly
contact to Mr.Seo.

---------- Forwarded message ----------
From: MAPRED'2010 <mapred2010@easychair.org>
Date: Wed, Sep 22, 2010 at 6:32 AM
Subject: MAPRED'2010 notification for paper 1 : ACCEPTED
To: "Edward J. Yoon" <edwardyoon@apache.org>


It is our pleasure to inform you that your paper HAMA: An Efficient
Matrix Computation with the MapReduce Framework has been ACCEPTED for
MAPRED'2010 workshop
at CLOUDCOM'2010.

Please, use this short time before the camera-ready submission to
improve your papers.

In particular, please consider comments from the most negative reviews.

We are looking forward to meeting you at the workshop.

Further details about final submission will come soon.


---------------------------- REVIEW 1 --------------------------
PAPER: 1
TITLE: HAMA: An Efficient Matrix Computation with the MapReduce Framework

OVERALL RATING: 2 (accept)
REVIEWER'S CONFIDENCE: 4 (expert)

The paper gives an overview of how Hama represents sparse and dense
matrices with column-storage Hbase and performs matrix computations
with mapreduce (multiplication and solving linear systems). Numerical
results with HAMA (using 2 different mapreduce implementations) are
compared with MPI.

The paper is well-written, but a bit light in detail, e.g. could gain
from providing a concrete example of the matrix representation (both
for sparse and dense) and more description of each field. More
background and explanation for the map() and reduce() methods
presented could also improve the paper (e.g. cancelled alternatives).

Regarding related work there seems to be a few missing publication
references, e.g.
a) Pregel: a system for large-scale graph processing
b) Distributed non-negative matrix factorization for dyadic data
analysis on mapreduce



---------------------------- REVIEW 2 --------------------------
PAPER: 1
TITLE: HAMA: An Efficient Matrix Computation with the MapReduce Framework

OVERALL RATING: 1 (weak accept)
REVIEWER'S CONFIDENCE: 3 (high)

This paper proposes a distributed framework designed for scientific
applications, which provides important primitives
such as matrix and graph computations. This framework, called HAMA, is
based on a layered architecture that makes
use of several computation engines, among which the MapReduce
framework for matrix computation tasks.
As a case study, the paper focuses on matrix multiplication and
solving linear equation systems. The HAMA approach
(built on top of MapReduce) is evaluated on 16 nodes and compared to
the MPI version of the same algorithms.

The paper is well organized and the matrix computation primitives are
clearly described. However, the authors could
also specify what other primitives are  provided by HAMA, as it is not
clear whether the framework supports only those
presented in the case study or it implements a wider range of matrix
computations.
Moreover, it is worth comparing the scalability of the HAMA approach
to the MPI implementation with respect to the
number of nodes used for the computation, not only as a function of
the size of the problem, as shown in the
experiments.
The paper does not include a related work section to compare the HAMA
framework to existing approaches that expose
computation primitives and it does not discuss the performance gain of
using the HAMA framework for scientific
applications.





-- 
Best Regards, Edward J. Yoon
edwardyoon@apache.org
http://blog.udanax.org

Mime
View raw message