Date: Mon, 1 Apr 1996 06:00:01 +1000 From: Bruce Evans <bde@zeta.org.au> To: andreas@knobel.gun.de, terry@lambert.org Cc: current@FreeBSD.org, j@uriah.heep.sax.de Subject: Re: FreeBSD performance vs BSD/OS Message-ID: <199603312000.GAA10850@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>If I remember right this was discussed already some months ago and >the result was, that freebsd has 128 HZ. BTW, I think it was Joerg, >whi told me that. HZ is not part of the application interface in BSD. >And if you run a benchmark like ssba (I think it was this one) that >tries to figure out HZ itself, then you get values about 128 or 130... >Perhaps someone else can make this more clear ?! I digged around in >the kernel sources in microtime and some other places, but didn't find >the right place... There are many different real and virtual (timekeeping) clocks with different frequencies: - the scheduling clock. This is a real clock with frequency that happens to be 100. It isn't available to applications. - the statistics clock. This is a real clock with frequency that happens to be 128. It isn't directly available to applications. - the clock reported by clock(3). This is a virtual clock with a frequency that happens to be 128. It's actual frequency is given by the macro CLOCKS_PER_SEC. Note that CLOCKS_PER_SEC may be floating point. Don't use clock() in new programs under FreeBSD. It is feeble compared with getrusage(). It is provided for ANSI conformance. It is implemented by calling getrusage() and throwing away information and resolution. - the clock reported by times(3). This is a virtual clock with a frequency that happens to be 128. It's actual frequency is given by the macro CLK_TCK (deprecated; don't use) and by sysconf(_SC_CLK_TCK) and by sysctl(3). Note that its frequency may be different fro CLOCKS_PER_SEC. Don't use times() in new programs under FreeBSD. It is feeble compared with gettimeofday() together with getrusage(). It is provided for POSIX conformance. It is implemented by calling gettimeofday() and getrusage() and throwing away information and resolution. - the profiling clock. This is a real clock with frequency 1024. It is used mainly by moncontrol(3), kgmon(8) and gprof(1). Applications should determine its actual frequency using sysctl(3) or by reading it from the header in the profiling data file. - the mc14618a clock. This is a real clock with a nominal frequency of 32768. It is divided down to give the statistic clock and the profiling clock. It isn't available to applications. - the microseconds clock. This is a virtual clock with frequency 1000000. It is used for most timekeeping in BSD and is exported to applications in getrusage(2), gettimeofday(2), select(2), getitimer(2), ... This is the clock that should normally be used by BSD applications. - the i8254 clock. This is a real clock with a nominal frequency of 1193182. It is divided down to give the scheduling clock. It isn't available to applications. - the i586 clock on i586 systems. This is a real clock with a frequency of up to 200000000. It is used to interpolate between values of the scheduling clock. It isn't available to applications. Summary: if "HZ" isn't 1000000 then the application is probably using the wrong clock. Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199603312000.GAA10850>