Circuit PSI with associated data benchmarking about volepsi HOT 5 CLOSED

It is supported but there is no command line tool to run it.

For the sender, you need to pass the associated data in here as the values parameter.
https://github.com/Visa-Research/volepsi/blob/main/volePSI/RsCpsi.h#L88

The resulting shared are written to here.
https://github.com/Visa-Research/volepsi/blob/main/volePSI/RsCpsi.h#L77
https://github.com/Visa-Research/volepsi/blob/main/volePSI/RsCpsi.h#L105

The receiver can also have associated values but the process works a bit differently. The receiver learns a 'mapping' that tell them which input value was mapped to position i. That is, position i will hold the input with index mMapping[i]. Note that mMapping will have 30% empty spots. They are denoted with mMapping[i]=-1. So for the receiver to give associated values, they can just add them after since they know the mapping.
https://github.com/Visa-Research/volepsi/blob/main/volePSI/RsCpsi.h#L108

Note that the sender also holds a mapping but this mapping is a bit ambiguous in that the sender knows that input with index i was mapped to one of the positions mMapping[i][0],mMapping[i][1],mMapping[i][2].
https://github.com/Visa-Research/volepsi/blob/main/volePSI/RsCpsi.h#L82

from volepsi.

1. you can not use the file based PSI command line interface with circuit PSI. This only works for normal PSI.

2. For circuit psi, you can only use silent ot and the default version of silent ot is with the Expand Accumulate code. Not with the faster but experimental Silver code.

You could modify the code to use IKNP or the better SoftSpoken OT protocol. This comes at a higher communication cost but better computational overhead.

3. as explained above, you can provide multiple associated values. You should just pack them together into the parameter oc::MatrixView<u8> values. There should be n rows and however many columns you want. each row is one associated values, so party A will have X and A_0, B_0, C_0 where element X[i] is associated with A_0[i], B_0[i], C_0[i]. You should then have values[i]=(A_0[i], B_0[i], C_0[i]).

4. for -perf -cpsi, all trails are added together and you get one value for the total running time https://github.com/Visa-Research/volepsi/blob/main/frontend/perf.cpp#L489

from volepsi.

I see, great! Thank you for your quick responses! Is there any example/instructions to invoke CPSI or do I need to call these two functions from my own C++ files?

from volepsi.

here's an example https://github.com/Visa-Research/volepsi/blob/main/tests/RsCpsi_Tests.cpp#L40

you have to write your own frontend code. Maybe something similar to this if you want it to run on the command line https://github.com/Visa-Research/volepsi/blob/main/volePSI/fileBased.cpp#L230

from volepsi.

Got it, thanks again! I'll follow the file-based PSI that has functionality separately for the sender and the receiver. When I get some time to clean it up I might send a PR :)

from volepsi.