To find out time elapsed for receiving a message at client's end in socket prg

Hi,

I want to calculate the time elapsed in receiving a message at client side. Here is what i am trying to do :
assuming buffer size for receiving is 500 bytes

timer start
1) Send a packet (say 500 bytes) from client to Server.
2) Server receives the packet and sends it back to the client.
3) Client Receives the packet back.'
timer stop
(Note that i am not concerned with how many steps server takes to send message back, that is not in our control since i am using TCP protocol)
So far I have tried following things.

1) time() function it basically gives time in seconds, so I have to discard it because here execution time should be in milliseconds, and I was always getting 0 seconds with this.
2)clock() function, it tells us the no. of cycles it takes for execution and then if u divide it by CLOCKS_PER_SEC, you can get the time elapsed, but in my scenario it was not working, always giving 0 seconds.
3) gettimeofday(&tv , NULL) : I used this and am getting time in microseconds, but my problem is , it is giving me weird results i.e sometimes I get 500 usec, sometimes 1000 usec, sometimes 5000 usec and sometimes 10000 usec also. I can' t trust so much of variations .

There is nothing much in code, it is just a client-server socket prg code implementing TCP protocol with loopback port.(127.0.0.1).

I hope I made myself clear in defining my problem , I want some good function, which can tell me the time elapsed in what i am trying to do.

thanks in advance

aditya

So, you are attempting to remake ICMP ping on a TCP socket? Sounds very suspiciously like homework.

In any case, if you want to calculate time, gettimeofday() is just fine. Don't know if clock() will work, since most of the stuff won't be controlled by the CPU, but the network card.

The numbers you provide don't sound that much weird to me. It is 0.5 - 10 msec. A +- 10 msec for a server/client type of code might not be strange

Quote:

---

Originally Posted by adityanahan

3) gettimeofday(&tv , NULL) : I used this and am getting time in microseconds, but my problem is , it is giving me weird results i.e sometimes I get 500 usec, sometimes 1000 usec, sometimes 5000 usec and sometimes 10000 usec also. I can' t trust so much of variations .

---

Post the code you are using for the timer. I would use a millisecond timer (eg, ntp_gettime if available) instead of a microsecond timer.

Also, there can easily be this much variation in times with networking.

I am posting part of the code from client side.

Client makes socket and connects to server
-----------------------------------------------------------------------------------------------------------------

Code:

---

gettimeofday(&tv1,NULL);
        if(send(sock,echoString,echoStringLen,0) != echoStringLen)  //sending starts here
                printf(" send failed ");
       
        totalBytesRcvd =0;
        printf("Received: ");
        int recv_count=0;
        while(totalBytesRcvd < echoStringLen)
        {
               
                if((bytesRcvd=recv(sock,echoBuffer,RCVBUFSIZE-1,0))<=0) //receiving from server
                  printf("received failed ");
                printf("bytes received = %d \n",bytesRcvd);
                  totalBytesRcvd += bytesRcvd;
                  echoBuffer[bytesRcvd]='\0';
                  puts(echoBuffer);
                recv_count++;       
        }
        gettimeofday(&tv2,NULL);
        totalT=(tv2.tv_sec-tv1.tv_sec)+(tv2.tv_usec-tv1.tv_usec);
              printf("recv count = %d and time taken = %ld microseconds \n",recv_count,totalT);

---

----------------------------------------------------------------------------------------------------------------------
thanks

Code:

---

totalT=(tv2.tv_sec-tv1.tv_sec)+(tv2.tv_usec-tv1.tv_usec);

---

This is a little wrong, but much of the time the result will be correct. If the transmission took 3 seconds, tho, using this equation that would only count as 3 microseconds.

Also, if the start is at 999999998 usecs and the end at 5 usecs, you will get a negative value. Plus one for the extra second :p Get it?

Presuming it will never actually be more than one second, you need to add a few more steps:

if (tv2.secs > tv1.secs) subtract 2.usecs from 1.usecs
else subtract 1 from 2

Make sense? This way you always have a positive number.

@MK27 thanks for ur response..

But I did not get what you are trying to say..Can u clarify in detail.

Code:

---

  This is a little wrong, but much of the time the result will be correct. If the transmission took 3 seconds, tho, using this equation that would only count as 3 microseconds.

---

How can a seconds be shown as microseconds.??I don't understand this.

Code:

---

Also, if the start is at 999999998 usecs and the end at 5 usecs, you will get a negative value. Plus one for the extra second  Get it?

---

Why are you saying this..I presume the tv2.secs should be always greater than tv1.secs .

I think I don't understand how gettimeofday works.. Can you explain me how it works.

thanks a lot

gettimeofday works like a clock with hours and minutes, but instead of hours and minutes it has seconds and microseconds. 4:05 is after 3:57, tho 5 is less than 57.

totalT is a count of microseconds. (tv2.tv_sec-tv1.tv_sec) is a count of seconds. You need to multiply that number by one million if you want to do it that way -- but don't, because then the second error in your equation will become much more drastic. As is, it is wrong, but it doesn't matter because at most it will add 1 usec to the count.

Actually what I wrote in that last post is a little wrong too,* it should have been:

if (tv2.secs > tv1.secs) subtract tv1.usecs from 1000000 then add tv2.usecs
else subtract tv1.usecs from tv2.usecs

Applying that to a normal clock (3:57 and 4:05), you would say 4 is greater than 3, so subract 57 from 60 (=3) then add 5. If it were 4:05 and 4:20, you would just subtract 5 from 20.

*and I used nanoseconds (999999998) instead of microseconds (999998), whoops.

I got your point, but I still have a doubt.

Since tv2 is after tv1, therefore tv2.sec > tv1.sec (this condition should be always true) Correct me if I am wrong on this.

So according to me we should always check tv.usec condition like
if(tv1.usec > tv2.usec)
1000000-tv1.usec + tv2.usec;
else
tv2.usec-tv1.usec ;

Wht do u think about this MK27

Quote:

---

Originally Posted by adityanahan

I got your point, but I still have a doubt.

Since tv2 is after tv1, therefore tv2.sec > tv1.sec (this condition should be always true) Correct me if I am wrong on this.

---

No. If it is 2:30, and five minutes go by, what hour is it? Still 2. But the minutes have changed.

hey MK27,

Thanks a lot. I figured out what I was doing wrong and resolved the issue.
basically to get the time elapsed, you have to multiply seconds by 1000000 and then add the microseconds to it.

time elapsed = (tv2.sec -tv1.sec)*1000000 + (tv2.usec-tv1.usec);

it doesn't matter if tv2.usec is smaller or greater than tv1.usec. Because tv2.sec will always be greater than tv1.sec so signs are taken care of by this statement.

I was looking into what exactly tv.sec returns and what it returns is # of seconds elapsed since Jan 1 ,1970. :)