Tassilo Horn wrote: > I've written a specialized compression utility (C++). Now I want to > compare it to some general purpose compressors like gzip, bzip2 and > compress. CPU time and memory consumption are the main things of > interest. > > Therefore I tried using GNU time [1] (version 1.7). With it I get the > correct CPU time, but the memory resources are always zero (%p and %M in > time's format string). I tested this on Linux 2.4 & 2.6, Mac OS X and > OpenBSD. > > Then I tried writing a simple replacement [2] which utilizes the > getrusage() (via wait4()) system call. But with this, the memory > resources are always zero, too. Looking into GNU time's source code, it > seems that it uses the same approach, so this was a no-go. ;-) > > Is getrusage()/wait4() broken, or not intended to be used like this? GNU libc uses mmap(MAP_ANONYMOUS) for large blocks; this isn't accounted for by conventional Unix resource handling mechanisms (getrusage, setrlimit etc). > How can I get correct memory resource usage informations of a (child) > process (without changing the code of the process' applications)? Parse /proc/<pid>/maps. Any entries with an inode of zero (other than the heap and stack regions) correspond to mmap(MAP_ANONYMOUS). -- Glynn Clements <glynn@xxxxxxxxxxxxxxxxxx> - : send the line "unsubscribe linux-c-programming" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
