Performance test (MPC and TEE)

[victormassy]

In the last PATCG meeting, a discussion was raised about the use of MPC or TEE for aggregating the reports. I would like to take part in this group by comparing performances of both MPC and TEE in the attribution reporting case. It seems to be a consensus that MPC is better for privacy, but I would like to measure the execution time on a large dataset.

I have been through the IPA research prototype. MP-SPDZ seems powerful. Do you think MP-SPDZ is suitable for performance test?

I would like to try different sorting algorithms. Do you know any other one than the one you are implementing?

[martinthomson]

Hi Victor, I think that this is a question better suited to the MP-SPDZ prototype in this repository.

[benjaminsavage]

There are many MPC sorting algorithms I'm familiar with.

• We've been implementing the one from this paper: https://eprint.iacr.org/2019/695.pdf and the maliciously secure variant of this protocol that's soon to be published.
• That paper references many other prior works that you can take a look at.
• We understand that many people are using Quicksort (a secure variant of it) that seems to work well in practice.
• I'm not sure exactly what sort algorithms are used for the default "sort" operations supported by frameworks like MP-SPDZ and Sharemind https://docs.sharemind.cyber.ee/2022.03/api/rmind-library