High Performance Graph Data Imputation on Multiple GPUs
Round 1
Reviewer 1 Report
The paper, although hard to read, addresses a real problem with high interest in the internet domain. It seems to me to be well designed and well structured, although I would prefer section 6 to be section 2. Then section 2 would become section 3 and so on.
The notation (sub-section 2.1) is quite incomplete and would be clearer if presented in a table. Authors must also standardize mathematical operators (e.g., sometimes use x other times use *).
Other aspects of detail to improve:
- 1: Please move Alg. 1 for after line 118
- Line 126: (.)+ does not appear in the previous expression of the same line????
- 2: Please move Alg. 2 for after line 212
- 6 to 8: Please move these figures for after line 348
- Line 343: Please replace “Sec. III.B” by “Sec. 3.2”
- 9: Please move this figure for after line 368
- 10: Please move this figure for after line 376
- Table 1: Please move this Table for after line 382.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper describes a parallel implementation of a graph data imputation algorithm.
The paper is well written and clearly understandable.
I have few comments for the authors to address prior publication.
1) Please, include the visual profiler results for the GPU baseline implementation and for the optimized implementation. These results can improve the quality of the paper. Moreover, an analysis of the computational weight of the various steps could help the reader in understanding sentences such as those at lines 192-196.
2) Is it possible to model how the parallel strategy works on systems with more than 2 GPUs? Modern supercomputing nodes are equipped with 4 or more GPUs.
Minor points and typos
- Please, do not use the capital letter after the colon.
- Please, rephrase lines 5-6.
- Line 112, please change “Here” with “here”.
- Line 317, please change “Turin” with “Turing”.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper presents a novel scheme to perform convolutional imputation algorithm implemented by multiple GPUs. The paper is very well written and organized. The topic addressed is timely and in the scope of the conference, the methodology followed is well described and sounds ok.
I have only two comments/recommendations:
- Figure 1: it would better to show a more detailed example to make it easier to understand
- Algorithm 1: line 5: what is NOTconverged? Is it a function call? Or is it "not converged"? (according to fig 2)
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors have addressed all the raised issues.
