Comparison with ag2pc

[nitincic]

emp-pvc appears to be almost as slow as ag2pc over 32 core LAN machines. Is it because of the more efficient multi-threaded implementation of ag2pc over pvc?

[fionser]

Hi, Nitin,
Could you provide some testing details ? For example, what circuit did you test ?

[nitincic]

I'm running a circuit for hashing (multiple sha256 pipelined). Initially pvc does slightly better but eventually, after up-sizing to about 64 sha256 blocks. ag2pc tends to outperform pvc (k=1/2).

[wangxiao1254]

This sounds strange to me. I would not expect this to happen. Xiao

…

On Fri, Jan 17, 2020 at 6:49 AM Nitin Agrawal ***@***.***> wrote: I'm running a circuit for hashing (multiple sha256 pipelined). Initially pvc does slightly better but eventually, after up-sizing to about 64 sha256 blocks. ag2pc tends to outperform pvc (k=1/2). — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#6?email_source=notifications&email_token=AARKGCWYUBPLDFUFP5EDNZTQ6GSOVA5CNFSM4KICI2ZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJHSRII#issuecomment-575613089>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AARKGCVI4CYEGELTXLMRPATQ6GSOVANCNFSM4KICI2ZA> .

[nitincic]

I can virtually even see 100s of threads running while executing the circuit with emp-ag2pc, while that's not the case with pvc.

[wangxiao1254]

The speed of emp-pvc should be similar to sh2pc. while ag2pc should be 7-10 times slower. The number of threads used in ag2pc is a tunable parameter (default it is 8). Xiao

…

On Fri, Jan 17, 2020 at 8:55 PM Nitin Agrawal ***@***.***> wrote: I can virtually even see 100s of threads running while executing the circuit with emp-ag2pc, while that's not the case with pvc. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#6?email_source=notifications&email_token=AARKGCX4WT3DY7N64J2OYVDQ6JVQ5A5CNFSM4KICI2ZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJJOX2Q#issuecomment-575859690>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AARKGCSTCTNJDQPJBWD3ZYLQ6JVQ5ANCNFSM4KICI2ZA> .

[fionser]

@nitincic

• By default, pvc uses (k/2) so we only launch two threads with two network connections to perform OTs and GC simulation, simultaneously.
• To change k, modify this line
• While a larger k will require to launch more threads and connections to run all the simulation in parallel.

Hope this help to solve your issue.