Skip to content

reduces the total RTT for revealOblivBoolArray - #67

Open
weikengchen wants to merge 1 commit into
samee:obliv-cfrom
weikengchen:patch-1
Open

weikengchen wants to merge 1 commit into
samee:obliv-cfrom
weikengchen:patch-1

Conversation

@weikengchen

@weikengchen weikengchen commented Jul 11, 2018

Copy link
Copy Markdown
Contributor

When we reveal data to both parties, we may use revealOblivBoolArray. It passes party=0 to the underlying function. There are many rounds due to underlying implementations.

This commit changes the way revealOblivBoolArray deals with an array to the same way used by revealObliv##tnameArray, which uses party=2 to reveal the result to the second party first, and then uses party=1 to reveal to the first party, rather than uses party=0.

It reduces the total RTT.

I tested the change with the following script:

#include<obliv.h>
#include <obliv.oh>
#include<stdio.h>
#include<stdlib.h>
#include<time.h>

double wallClock()
{
  struct timespec t;
  clock_gettime(CLOCK_REALTIME,&t);
  return t.tv_sec+1e-9*t.tv_nsec;
}

void ocTestUtilTcpOrDie(ProtocolDesc* pd,const char* remote_host,
                        const char* port)
{
  if(!remote_host)
  { if(protocolAcceptTcp2P(pd,port)!=0)
    { fprintf(stderr,"TCP accept failed\n");
      exit(1);
    }
  }
  else 
    if(protocolConnectTcp2P(pd,remote_host,port)!=0) 
    { fprintf(stderr,"TCP connect failed\n");
      exit(1);
    }
}


double lap;

#define MAXN 1000
void readBool(obliv bool* dest, int n, const bool* src,int party)
{
	OblivInputs specs[MAXN];
	int i;
	for(i=0;i<n;++i) setupOblivBool(specs+i,dest+i,src[i]);
	feedOblivInputs(specs,n,party);
}

void gotest(void* vargs)
{
	obliv bool p1_input[1000];
	obliv bool p2_input;

	bool buf[1000];
	readBool(p1_input, 1000, buf, 1);
	readBool(&p2_input, 1, buf, 2);

	fprintf(stderr,"OT time: %lf s\n",wallClock()-lap);

	for(int i = 0; i < 1000; i++){
		p1_input[i] = p1_input[i] ^ p2_input;
	}
        // no party knows

        // original version
        // for(int i = 0; i < 1000; ++i) revealOblivBool(buf + i, p1_input[i], 0);

        // improved version
	revealOblivBoolArray(buf, p1_input, 1000, 0);
}

int main(int argc,char* argv[])
{ 
  ProtocolDesc pd;

  const char* remote_host = (strcmp(argv[2],"--")==0?NULL:argv[2]);
  int i, party = (!remote_host?1:2);

  ocTestUtilTcpOrDie(&pd,remote_host,argv[1]);
  setCurrentParty(&pd,party);

  lap = wallClock();
  execYaoProtocol(&pd, gotest, NULL);
  fprintf(stderr,"Total time: %lf s\n",wallClock()-lap);
  cleanupProtocol(&pd);
  fprintf(stderr,"\n");
  return 0;
}

The original version:

OT time: 0.628520 s
Total time: 22.787672 s

The improved version:

OT time: 0.586414 s
Total time: 0.586716 s

@weikengchen

weikengchen commented Jul 11, 2018

Copy link
Copy Markdown
Contributor Author

The example in #66 considers 12345 bools.
I can't help but do an experiment for 12345.

The original version:

OT time: 0.643174 s
Total time: 255.182363 s

The improved version:

OT time: 0.652290 s
Total time: 0.655129 s

For this specific and boring application (1-bit-input from the evaluator, a lot of XOR free gates, 12345-bit-output), the performance is improved by 390x.

The example changes part of the code in the post into the following:

#define MAXN 12345
void readBool(obliv bool* dest, int n, const bool* src,int party)
{
        OblivInputs specs[MAXN];
        int i;
        for(i=0;i<n;++i) setupOblivBool(specs+i,dest+i,src[i]);
        feedOblivInputs(specs,n,party);
}

void gotest(void* vargs)
{
        obliv bool p1_input[12345];
        obliv bool p2_input;

        bool buf[12345];
        readBool(p1_input, 12345, buf, 1);
        readBool(&p2_input, 1, buf, 2);

        fprintf(stderr,"OT time: %lf s\n",wallClock()-lap);

        for(int i = 0; i < 12345; i++){
                p1_input[i] = p1_input[i] ^ p2_input;
        }

        for(int i = 0; i < 12345; ++i) revealOblivBool(buf + i, p1_input[i], 0);
        //revealOblivBoolArray(buf, p1_input, 12345, 0);
}

@weikengchen weikengchen changed the title reduces the RTT for revealOblivBoolArray reduces the total RTT for revealOblivBoolArray Jul 11, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant