Preprocessing prime is not compatible

[jiaaaarui]

This is a follow-up to this issue

I applied the commit patch and attempted again. I ran ./mascot-offline.x -p 0/1 -N 2 -P 115792089210356248762697446949407573530086143415290314195533631308867097853951 export-a2b on two terminals and got the output

115792089210356248762697446949407573530086143415290314195533631308867097853951 is not compatible with our encryption scheme, must be 1 modulo 32768

is this output expected? The preprocessing works with the fake offline, which leads me to think it should work here as well.

[mkskeller]

b47c9bb should fix this.

[jiaaaarui]

Thank you for your response. However, when I run it now, it seems that no edabits are being created in preprocessesing. I can provide more details later.

[jiaaaarui]

I am running ./mascot-offline.x -p 0/1 -N 2 -P 115792089210356248762697446949407573530086143415290314195533631308867097853951 export-a2b as before. I then run ./mascot-party.x -F -P 115792089210356248762697446949407573530086143415290314195533631308867097853951 -N 2 0/1 exp ort-b2a which gives error message

Fatal error at export-b2a-b2a-1:118 (DABIT): Something is wrong with the preprocessing data on disk: IO problem when buffering SPDZ gfp daBits from Player-Data//2-p-256/daBits-p-P0
Have you run the right program for generating it, such as './Fake-Offline.x 2 -P 115792089210356248762697446949407573530086143415290314195533631308867097853951'?

I checked the Player-Data directory, and it returns

$ ls Player-Data/2-p-256/
Opens-p-P0  Params-Data           Player-MAC-Keys-p-P1
Opens-p-P1  Player-MAC-Keys-p-P0

is there a special argument i need to create edabits when invoking mascot-party.x?

[mkskeller]

It looks like you're running the offline phase for export-a2b but the online phase for export-b2a. Nevertheless, the offline phase should produce some edaBits in the Player-Data directory. What commit are you using?

[jiaaaarui]

I am on an older committ but with the patch applied. I will try again on a fresh commit and see how it fairs.

[jiaaaarui]

Hi, sorry about the delay. I just tested this on a the newest commit and the behavior is the same

[jiaaaarui]

This is the outputs in player-data

[mkskeller]

What is the output of Scripts/prep-usage.sh export-a2b or Scripts/prep-usage.sh export-b2a, depending on which one you've ran? On my end, I do get files for daBits being produced with export-b2a, and edaBits being produced with export-a2b. It might also help to start with a clean clone of the repository.

[jiaaaarui]

This is my output,

(base) garysong@ubuntu:~/clean/MP-SPDZ$ ./Scripts/prep-usage.py export-a2b
        3200 bit opens
      422270 bit triples
          30 integer opens
         400 strict edabits of length 256

I just checked on a clean install and this is what my terminal fee looks like

(base) garysong@ubuntu:~/clean/MP-SPDZ$ ./compile.py -E mascot export-a2b --prime 1157
92089210356248762697446949407573530086143415290314195533631308867097853951 
WARNING: --prime/-P activates code that usually isn't the most efficient variant. Consider using --field/-F and set the prime only during the actual computation.
The chosen prime is particularly inefficient. Consider using a prime that is closer to a power of two, for example, 115792089237316195423570985008687907853269984665640564039457584007913126920193.
Use '-- --prime <prime>' to specify the prime for execution only.
Default bit length for compilation: 254
Default security parameter for compilation: 40
Compiling file Programs/Source/export-a2b.py
WARNING: No option selected for A2B conversion, defaulting to edaBits. Use -X/-Y/-Z to get rid of this warning.
WARNING: Order of memory instructions not preserved, errors possible
Writing to Programs/Bytecode/export-a2b-a2b-1.bc
Writing to Programs/Functions/a2b-s:10-sbv:[10x16]
Warning: read before write at register s0(10)/7ffffe4fe600
        register trace: <omitted>
Writing to Programs/Schedules/export-a2b.sch
Writing to Programs/Bytecode/export-a2b-0.bc
Hash: 1f9af4ae659e3d57719bb2a94d38ebb331d72ba4628176539b6bd8a900dfcac3
Program requires at most:
          30 integer opens
      422270 bit triples
         400 strict edabits of length 256
        3200 bit opens
        3634 virtual machine rounds
Expected communication is 0.219455 MB online and 236.471 MB offline.
(base) garysong@ubuntu:~/clean/MP-SPDZ$ ./mascot-offline.x -p 1 -N 2 -P 1157920
89210356248762697446949407573530086143415290314195533631308867097853951 export-
a2b
Using statistical security parameter 40
Writing to file in Player-Data//2-p-256/
Significant amount of unused triples of Tinier distorting the benchmark. This protocol has a large minimum batch size, which makes this unavoidable for small programs.

The only line I ran before was make setup and make mascot-offline.x

[mkskeller]

Thank you for this. I realised that I couldn't reproduce it because I missed the -P parameter in the compilation. You should find that 6cac6d1 fixes the bug.

[jiaaaarui]

Thank you for the update. On my clean repo, I pulled the update and recompiled mascot-offline.x and ran it as such. I got the following error

(base) garysong@ubuntu:~/clean/MP-SPDZ$ make mascot-offline.x -j8
clang++ -o Machines/mascot-offline.o Machines/mascot-offline.cpp -DNO_AVX_OT -I./local/include -march=native  -g -Wextra -Wall -O3 -I. -I./deps -pthread    -DUSE_GF2N_LONG '-DPREP_DIR="Player-Data/"' '-DSSL_DIR="Player-Data/"'  -std=c++17 -Werror  -std=c++17 -fPIC -MMD -MP -c
clang++ -o mascot-offline.x -DNO_AVX_OT -I./local/include -march=native  -g -Wextra -Wall -O3 -I. -I./deps -pthread    -DUSE_GF2N_LONG '-DPREP_DIR="Player-Data/"' '-DSSL_DIR="Player-Data/"'  -std=c++17 -Werror  -std=c++17 -fPIC Machines/mascot-offline.o libSPDZ.so Machines/Tinier.o OT/BaseOT.o OT/BitDiagonal.o OT/MascotParams.o OT/OTExtension.o OT/OTExtensionWithMatrix.o OT/OTTripleSetup.o deps/SimplestOT_C/ref10/libSimplestOT.a local/lib/liblibOTe.a local/lib/libcryptoTools.a -lgmpxx -lgmp -lsodium   -Wl,-rpath -Wl,/home/garysong/clean/MP-SPDZ/local/lib -L/home/garysong/clean/MP-SPDZ/local/lib -lssl -lcrypto -lboost_filesystem -lboost_iostreams -lrt -z noexecstack -Wl,-rpath -Wl,/home/garysong/clean/MP-SPDZ li
(base) garysong@ubuntu:~/clean/MP-SPDZ$ ./mascot-offline.x -p 0 -N 2 -P 11579208
9210356248762697446949407573530086143415290314195533631308867097853951 export-a2
b
Using statistical security parameter 40
Writing to file in Player-Data//2-p-256/
106427501127946520196583018678435888206051970758723569473975998785679572758166 != 106427510833527319868645205648480045976446072337976114222431774392618259737897
bits: 0110111000110110010101000100100001010000110101100001010010010101001001000101001101110111101100100101011101101001000110011110100100100111101110111000101100111000110110101111000110001011101110110101101001101111111010100011101100100100110010111101001011010111
terminate called after throwing an instance of 'Offline_Check_Error'
  what():  Offline-Check-Error : edabit shuffle opening
Aborted

[mkskeller]

What is your setup like processor and the versions of OS and clang? Can you build the attached Dockerfile.txt?

[jiaaaarui]

Hi, I think I fixed the issue by adding arch= to CONFIG.mine. On another note, it seems that mascot-offline.x doesn't have an option for non-montgommery form like fake-offline.x does. is this intentional?

[mkskeller]

I'm not sure ARCH= is a good solution because it falls back onto the lowest capabilities of the processor family. mascot-offline.x does not have an option to disable Montgomery because the online phase binaries don't have it either.

[jiaaaarui]

I think this issue might be resolved since it arose on orbstack but not on WSL. I will have to check more in depth