govern@houxf.UUCP (07/14/83)
On my 4.1BSD system, the output of "time" for a long, CPU-intensive program typically shows a "real time" about 20 times as high as the CPU time, and the "user" time is often close to the real time. 1) Is this normal? The ratio seems awfully high. How can I check to see if the system is running ok or if it needs tuning? 2) Exactly what times go into the CPU and user times? - When a process has 20% of the CPU, is the "CPU time" value scaled to reflect this? - Where does IO time fit in? - What about paging time? (The application was a very large simulation, so it has to page a lot.) Thanks; Bill Stewart houxf!govern ucbvax!ihnp4!houxf!govern allegra!houxf!govern
drockwel@bbn-vax@sri-unix.UUCP (07/16/83)
From: Dennis Rockwell <drockwel@bbn-vax> Your question indicates a basic misapprehension: both the "user time" and the "sys time" given by the time command are CPU times. The user time is that time spent in your program, and the sys time is the time the operating system spent supporting your program. Thus, if your "user time" is close to the "real time", then your system is pretty well tuned, at least for CPU-bound processes. Of course, the terms I used above are subject to quantization errors; to be more accurate, the CPU times are the proportion of the times that the clock ticked while your program had control of the CPU. IO time gets added in (as sys time); also, if the clock ticks while the system is servicing a device interrupt, that time gets added to yours, even if the device that interrupted has nothing to do with you.