Secure Genomic String Search with Parallel Homomorphic Encryption

Round 1

Reviewer 1 Report

The paper describes a Fully Homomorphic Encryption (FHE) computation framework enablking the execution of string search operations with different applications and showing some improvements over state of the art.

The contribution is well organized and sufficiently clear. Authos shold consider the following points:

-authors say they extend a previous work, the extension should be detailed over [17]

-Table 1 in the introductiuon reports comparison with a framework, but comparison with other works based on GPU acceleration should be provided as done in table 3

Author Response

Thanks for reviewing the paper. We have edited Table 1 and the introduction to accommodate the changes asked on the reviews

Reviewer 2 Report

The manuscript is ready for publication. I have no comments.

NA

Author Response

Thank you for reviewing the paper

Reviewer 3 Report

Secure Genomic String Search with Parallel Homomorphic Encryption

1. In this paper, authors extend their work and apply it to secure gene search. This paper cannot outperform cuFHE when comparing gate operations but is 3.9 times faster on the arithmetic circuit.

2. Overall, this paper provides details on how they accelerate TFHE with CPU and GPU parallelism, but the difference is unclear compared to prior works. Especially, the GitHub code points to a 4-year-old link, which seems to be the artifact of [17] rather than this paper. Moreover, the author is based on a 6-year-old TFHE source, even though there are some recent TFHE versions (ex. concrete). Modify from a more recent version should achieve a better performance.

3. The depiction is not clear enough:

   1. In line 264, the author coalesces n ciphertext with length m into a 1xmn vector. But no

      explanation in Figure 3 on how 𝐴⃗!"#, ... , 𝐴⃗$ and 𝐵!"#, . . . , 𝐵$ is converted. Same issue in

      the Figure 2.

   2. The CPU II and GPU II parallelism is mentioned interleaved which might confuse the

      reader and the explanation tends to be fragmentary and lacks systematized.

   3. Need more explanation in Figure 6. Not able to know if the score is Tseq or Tpar without

      the axis description.

4. Experiment:

   a. Since this work extends from the [17], it should include the comparison to [17]
   b. GPU is not as accurate as CPU since it is designed for relatively low-precision calculation.

   Need to add some explanation or experiment to compare the output difference after

   moving the tasks to GPU.

5. Some minor issues:

   a. Lines 84-85 look duplicate to lines 77-78.
   b. Line 121/139 refers to the Table. 8 on page 20, which is too far away.

Minor editing of English language required

Author Response

We would like to thank the reviewer for the insightful comments. The paper has been revised to address and accommodate some of the comments at best of our abilities. 

• We have added some of the recent works (i.e., concrete) in the related works section.
• We updated the depictions and some of the symbols in the latest submissions
• We have added a new discussion around the accuracy of the computation around GPU operations
• The minor issues as highlighted in the review are also edited.

Round 2

Reviewer 3 Report

Information about References [13-15] is not sufficient. Please at least include their web addresses in the revision.

Author Response

Thanks for noticing that the links are missing which was due to a Latex bibliography error. Added them in the latest version.

Author Response File: Author Response.pdf