Subject: Referees Report C467 Resent-Date: Fri, 21 Jul 2000 09:12:11 -0400 Resent-From: Geoffrey Fox Resent-To: Geoffrey Fox Date: Fri, 21 Jul 2000 13:55:44 +0100 From: Juri Papay Organization: University of Southampton To: fox@csit.fsu.edu, jp@ecs.soton.ac.uk Dear Professor, I am sending you the referee's report of C467: VGDS: A Distributed Data Structure Framework for Scientific Computation by Pangfeng Liu and Jan-Jan Wu With Regards Juri Papay ------------------------------------------------------------------------------ CandC:PandE Referee Report Form CandC:PandE Referee Report Form -- Electronic Transimission to fox@csit.fsu.edu strongly preferred Referees Home Page: http://aspen.csit.fsu.edu/CandCPandE/ Email fox@csit.fsu.edu for URL of full paper to be reviewed WILEY Journal Home Page John Wiley and Sons, Ltd. Baffins Lane, Chichester West Sussex, PO19 1UD, England Telephone: (01243) 779777 Fax: (01243) 770379 REFEREE'S REPORT Concurrency and Computation:Practice and Experience A: General Information Please return to: Geoffrey C. Fox Electronically Preferred fox@csit.fsu.edu Concurrency and Computation: Practice and Experience Computational Science and Information Technology Florida State University 400 Dirac Science Library Tallahassee Florida 32306-4130 Office FAX 850-644-0098 Office Phone 850-644-4587 but best is cell phone 3152546387 Please fill in Summary Conclusions (Sec. C) and details as appropriate in Secs. D, E and F. B: Refereeing Philosophy We encourage a broad range of readers and contributors. Please judge papers on their technical merit and separate comments on this from those on style and approach. Keep in mind the strong practical orientation that we are trying to give the journal. Note that the forms attached provide separate paper for comments that you wish only the editor to see and those that both the editor and author receive. Your identity will of course not be revealed to the author. C467: VGDS: A Distributed Data Structure Framework for Scientific Computation by Pangfeng Liu and Jan-Jan Wu C: Paper and Referee Metadata Paper Number Cnnn: C467 Date: 20.07.2000 Paper Title: VGDS: A Distributed Data Structure Framework for Scientific Computation Author(s): Pangfeng Liu and Jan-Jan Wu Referee: Juri Papay Address: Department of Electronics and Computer Science, University of Southampton, Southampton SO17 1BJ, UK, Tel: +44 (0)2380 592369, Email: jp@ecs.soton.ac.uk Referee Recommendations. Please indicate overall recommendations here, and details in following sections. reject D: Referee Comments (For Editor Only) This paper requires substantial rewriting and I cannot recommend it for publication. E: Referee Comments (For Author and Editor) 1. The paper is poorly structured and it is difficult the read and understand. 2. The English should be thoroughly checked, there are numerous typos and complicated sentences which require rephrasing. 3. The results are poorly presented. F: Presentation Changes 1. Introduction The lack of scalable distributed data structures in C++ is not the only reason for the slow acceptance of C++ by the HPC community, maybe the authors need to give a more balanced view. In the introduction the authors mention that the performance issue of distributed data structures needs to be addressed, however there is no evidence that this problem was seriously investigated. The measurements obtained for 4 workstations mentioned in section 5 provide little information about the performance and scalability of distributed data structures. 2. VGS framework a.) Formatting Table 1, move the caption. b.) Section 2.3. Data Decomposition Classes The global data structure are... (change to is ) c.) Section 2.3. Communication Classes Rephrase the following sentence: " Currently VGDS supports a message abstraction that flattens the attributes and contents of a data structure section to be communicated into a liner buffer on the sending side..." Explanation for incremental data redistribution required. " During the course of computation, if the VGDS data object detects that a remapping is necessary (e.g. for load balancing purpose), it invokes the remap method in Mapper, which in turn redistributes the data structure incrementally." 3. Implementation issues a.) Section 3.1 Move Figure 5 to the relevant paragraph in the text i.e. to page 8. The data distribution for regular arrays is missing from Figure 5, although the following sentence from page 8 refers to it. "Figure 5 shows the duplication mechanism for a regular array, an unstructured mesh, and an adaptive Barnes-Hut tree for N-body algorithms." b.) Section 3.2. Instead of "Optimisatin for Good Performance" use "Performance Optimisation" There is no need to split this section into four sub-sections. c.) Section 3.2.1. "Block distribution is oftenly..." change to often used. Typing error procerssors "Each ORB node is assigned a weight equal to the total number of operations performed in updating the states of elements over all its descendant procerssors (processors)." Typing error "At each steop (step) of the load-balancing process it is necessary to move bodies from the overloaded child to the non-overloaded child." d.) Section 3.2.3 Typing error in "traveral". e.) Section 3.2.4. "The communication style in computation..." change to communication pattern. Rephrase paragraph starting with "VGDS employs a randomized scheduling ..." 4. Case Study:BH Tree a.) 4.3. Parallel Implementation What is self-contained? "To make the paper self-contained, we briefly describe our parallel implementation of the BH algorithm, upon which the BH-Tree library is built." Reformat text in Figure 8, line spacing between (b) and (c). b.) 4.4. The Tree Framework Typing error "frameowrk" "To eliminate duplicated programming investments in developing similar tree-based scientific codes, we have developed a VGDS tree frameowrk." c.) Section 4.4.1 Base tree layer Rephrase the paragraph starting with: " Tree reduction computes the data of a tree node according..." d.) Section 4.4.2 Barnes-Hut tree Rephrase the following sentence: "The BH_tree layer supports tree operations required ..." Rephrase the following sentence: "The computation class Interaction (Figure 12) goes through the essential data list and calls for functions to compute body-to-body and body-to-cluster interactions defined by the user of Interaction." e.) Section 4.4.4 What is class 15, typo error in orthogonal. "The ORBPartitioner class 15 inherits the Mapper class and de_nes functions that are specialized for the orthogoal (orthogonal) recursive bisection partitioning method." Repetition from 3.2.4 "The communication phases in tree codes and unstructured mesh codes..." Repetition from 3.2.4 " VGDS employs a randomized protocol for all-to-some communication..." 5. Experimental results Figure 14 is not referenced in this section. Table 2 is in Related work section. More detailed elaboration of the following statement is required: "The VGDS class libraries significantly reduced the code sizes (e.g. for each of the two tree codes, from over ten thousand lines down to a few hundred lines) and the development time (from over six months down to a few days) of these applications, compared with their message-passing C counterparts." Section 5 is very short and does not present enough experimental data to highlight the efficiency of distributed data structures. This is a core section of the paper and it requires significant rework. 6. Related work This section needs rewriting. Figure 15 is not referenced, move Figure 15 and Table 2 to a different section. Rephrase the following sentence: "In a related work by ourselves [6], we report abstractions of adaptive load balancing mechanisms and complex, many-to-many communications as C++ classes for supporting HPC challenging applications." Typing error in matrix "Instead of tackling one particular data structure such as arrays or matrics ..." Missing reference "This looks very much like the structure of the Nexus run-time system which provides support for CC++ [] where multiple threads are mapped onto a context (Vnode in POOMA) and multiple contexts are mapped onto a physical node." Clarify the following sentence "Indeed, our hand-written prototypes substantially outperform all the efforts mentioned, while allowing greater generality." 7.Conclusion Can you provide some data to support the following statement "The penalty due to object orientation is about 10% compared with its C-only implementation." Typing error in reducing "We are currently investigating possible approaches to reduing such overhead."