Here is the composite review of the GASS paper -- note, I didn't receive a couple. Paper Title: GASS: A Data Movement and Access Service for Wide Area Computing Systems First Author: Joseph Bester Referee Name: CS660 composite class review 1) Interest of the Subject (not the paper) to the conference: Scores: 4 5 5 4 5 5 Average: 4.67 Std: 0.52 2) Confidence in your REVIEW: Scores: 3 3 3 3 4 3 2 3 Average: 3.00 Std: 0.53 3) Contribution of the paper: Scores: 4 4 4 3 4 5 3 4 Average: 3.88 Std: 0.64 4) Technical content: Scores: 3 4 4 4 4 5 3 3 Average: 3.75 Std: 0.71 5) Presentation style: Scores: 4 4 5 3 4 5 4 4 Average: 4.12 Std: 0.64 Overall Recommendation: (circle one) Strong Accept is 5, Strong Reject is 0 Scores: 3 4 4 3 4 4 2 4 Average: 3.50 Std: 0.76 2) Interest of the subject (not the paper) to the conference Scores: 4 5 5 4 5 5 Average: 4.67 Std: 0.52 3) Novelty of the problem: Scores: 2 4 3 2 3 2 Average: 2.67 Std: 0.82 4) Originality of the solution: Scores: 3 3 4 2 4 5 2 2 Average: 3.12 Std: 1.13 5) Expected impact of the results in the paper: Scores: 4 3 4 3 4 5 3 Average: 3.71 Std: 0.76 ------------------------------------------------------------ A nice three line summary from one of your reviews: The authors provide thorough introduction to the need for distributed computing and to the diverse tools available, with hard sell for GASS. The performance study gives a clear picture of overheads involved, but data analysis is skimpy. Some quotes from your reviews: ------------------------------------------------------------ 1. Please comment on the original contributions of this paper. > The GASS implementation and relevance to grid computing are laid > out nicely. This document provides a good indication of how the > Globus team is addressing the need for secondary storage within > an application. While the methods used by GASS for cache management, > replication, etc, do not appear to be ground-breaking technology, > they rely on technologies that have matured to a level where they > can be quickly integrated into a production environment with a high > level of confidence. > The URL way to store data in a shared > computing environment is to my knowlege, a novelty, and a convenient > enhancement. Unfortunately, in the actual stage of the research > and according to the speed tables, I don't see the advantage to use > it. > I thought the possiblity of easily downloading an executable and > simply running was good as well as viewing URL references as simple > file locations and being able to cache those names. ------------------------------------------------------------ 2. Please comment on the significance of the paper. What aspects of the paper are likely to be used by other researchers and/or practitioners? > I think this is just one of many distributed computing tools. > If it is easier to use than others, people will use it. > > Unfortunately, I do not think that these tools are judged on > their real merits, because no impartial researcher [ whose salary > does not depend on GLOBUS or CONDOR or whatever ] has the time > to run his project multilple times with different tools and > then compare results. The only people who can do something like > that are doing it so they can sell their own project. > I think this presents an alternate to current remote file access > protocols. In the future, I would like to see researchers > develop the same type of system, but for them to be self-contained. > This system requires the use of globus. > Also, I think the subject of file coherency could be more > fully explored. ------------------------------------------------------------ 3. Please list any technical errors, uncited related work, and/or ways in which the technical material could be improved. >Many external references without substantiation to the Globus system. >As for errors in the paper, I would like to have seen the basis for their >conclusions in the "Common Grid File Access Patterns" section. The >conclusions drawn in this section are the basis for the entire GASS project, >yet no reference to previous work is present. Section 6 needs to be >refined, the experimental results were difficult to interpret and not >visually impressive. A few well commented graphs could improve this section >dramatically. Amen. > I would have like to have seen one complete example [ code + > platform information + results ] or a reference to a complete > example. The example at the end of the article was good, but > lacked code. > Also, I was not convinced that the types of file access scenarios > provided by GASS meet the demands of most applications. This would > have been easier to believe had a survey been performed. In short, I > am not certain that GASS is adequately building upon previous work. > I am still concerned about the use of the API's, considering the > use of the words "programmer transparency". ------------------------------------------------------------ 4. Please add any additional comments for the authors (e.g., how might the paper be further improved beyond 1-3 above?). > I would like to see if GASS could be used to access files while they > are being transferred to the local cache, i.e. streaming > simultaneously to both the cache and to the process which needs the > data. >To fully appreciate the effect of GASS on real applications, the >authors need to present one or more case studies that show the gains (if >any) achieved through th euse of GASS vs using other techniques for data >movement in GLOBUS (if any existed prior to GASS). >The explanation for the GASS implementation in section (4.2) is >inadequate in providing an idea about the services provided by the GASS >servers and how to make use of such services. >In section (6) the authors make reference to x-gass URL's without >defining this term. >In table (2), it is noted that certain tests in the third part >(Prestaged file read from /tmp with GASS mechanisms) give better >performanc ethan their counterparts in the first part (File read from /tmp >without GASS mechanism), the authors do not give any explanations as to >why this anomaly occured. > Even though the whole GASS concept is a priori interesting, > according to Table 1 and 2, I don't see the actual advantage, > besides, maybe, shorter setup-time for distributed applications. > Paragraph 5.3, is talking about a world record, how much was the > GASS contribution involved in it? > Again like RIO VISTA, it would be interesting to have comparaisons > between GASS and other systems (more than No Cache, or To memory)