It seems that while top lists kernel provided statistics per process which is somewhat interesting but not all that useful, perf is really sampling the system, and gives a real picture of who's hogging your system, which is usually why you've started top in the first place.
Let me give a trivial example. If your system is thrashing, top will not give you a clue about what's bothering your system. It'll show processes coming and going from the top ten, with no apparent reason. perf top, should show consistently a very high count on a *pageout* related kernel function. You don't have to look at the load Likewise, a system with fork bomb, the relevant functions should show up, and make it clear what the root of your problem is. While I haven't actually tried it for complicated situations, it looks like a similar tool for windows works very well[0] (note, this is not exactly the same situation. Does anyone have experience with perf top? Is it really working better than top/htop? [0] e.g. http://randomascii.wordpress.com/2012/09/04/windows-slowdown-investigated-and-identified/, yes, it's not 100% the same thing, because he also recorded full stack traces, so it's more like perf record -a, but it's pretty close, and you can see it's working well enough even before looking at the full stacktrace.
_______________________________________________ Linux-il mailing list Linux-il@cs.huji.ac.il http://mailman.cs.huji.ac.il/mailman/listinfo/linux-il
