We want to find out the time elapsed between a disk I/O request generated and it being serviced.

We tried to do so by using the difference between jiffies but many disk I/O requests are serviced before a "jiffy" is elapsed. So the difference returned 0.

Another approach is to use do_gettimeofday() function from time.h but we feel that the overhead in finding the time would be pretty huge if this function is fired for each request.

Please hint us on how to proceed with this so as to optimize the process.

We want to find this time because we need to monitor the workload on a system.

well why don't you measure the same time interval for many disk accesses and then try to divide to get the time per single access? :P

have you tried using the time stamp counters ?
http://lwn.net/Articles/209101/

Thanks a lot for the replies.. Will look into it and post the result (if successful) here.

struct request {

sector_t __sector; /* sector cursor */

unsigned long start_time;

int ref_count;

unsigned long deadline;

/*and many more*/

};

Could anyone explain what each of these members of struct request (/include/linux/blkdev.h) mean ?