Comments on: Why can’t VTune be more like Shark? http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/ Words Wed, 07 Jan 2009 06:10:55 +0000 http://wordpress.org/?v=2.6.5 By: Bluehive http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-3080 Bluehive Thu, 06 Nov 2008 12:34:02 +0000 http://blog.vlad1.com/?p=61#comment-3080 Hey Vladimir, I'm quite late to post a reply as I happen to see this interesting blog today itself. I'd like you to see one more tool HP Caliper. Take my word, you'll get addicted to it like me :). Besides HP-UX, it works on Linux as well. The call graph feature you mentioned is available with Caliper in two manner: 1. sampled call graph 2. Sampled call Stack. It can give you every info you need including thread primitive data, application blocking/running status, dis-assembly listing and so on. I know Shark is good but Caliper is a serious contender :-) Only downside is that it's available on Intel Itanium platform only. Bluehive Hey Vladimir,
I’m quite late to post a reply as I happen to see this interesting blog today itself. I’d like you to see one more tool HP Caliper. Take my word, you’ll get addicted to it like me :). Besides HP-UX, it works on Linux as well. The call graph feature you mentioned is available with Caliper in two manner: 1. sampled call graph 2. Sampled call Stack. It can give you every info you need including thread primitive data, application blocking/running status, dis-assembly listing and so on. I know Shark is good but Caliper is a serious contender :-) Only downside is that it’s available on Intel Itanium platform only.

Bluehive

]]> By: BSBlog » Blog Archive » Profiling Dromaeo Testcases with Shark http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2895 BSBlog » Blog Archive » Profiling Dromaeo Testcases with Shark Fri, 05 Sep 2008 01:22:41 +0000 http://blog.vlad1.com/?p=61#comment-2895 [...] Shark should do a little dance and pop up a profile viewer. For a quick overview on using the Shark profile viewer, see Vlad’s blog. [...] [...] Shark should do a little dance and pop up a profile viewer. For a quick overview on using the Shark profile viewer, see Vlad’s blog. [...]

]]> By: vladimir http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2310 vladimir Sun, 20 Jul 2008 22:54:54 +0000 http://blog.vlad1.com/?p=61#comment-2310 @roc: I think AQtime has changed significantly since then; I just wrote up a post at http://blog.vlad1.com/2008/07/20/aqtime/ about my quick eval of it. I'll be talking about its allocation profiler as well, since it seems pretty useful. @roc: I think AQtime has changed significantly since then; I just wrote up a post at http://blog.vlad1.com/2008/07/20/aqtime/ about my quick eval of it. I’ll be talking about its allocation profiler as well, since it seems pretty useful.

]]> By: AQtime at Vladimir Vukićević http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2309 AQtime at Vladimir Vukićević Sun, 20 Jul 2008 22:52:08 +0000 http://blog.vlad1.com/?p=61#comment-2309 [...] About « Why can’t VTune be more like Shark? [...] [...] About « Why can’t VTune be more like Shark? [...]

]]> By: Robert O'Callahan http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2308 Robert O'Callahan Sun, 20 Jul 2008 22:27:12 +0000 http://blog.vlad1.com/?p=61#comment-2308 I tried AQTime a while back. It sucked, I think because it didn't do call-stack sampling, which is what Shark does and what VTune doesn't (which is the fundamental reason why VTune sucks, although the UI is horrible too). Sysprof and oprofile (and dtrace?) do call-stack sampling. Oprofile doesn't really have UI, sysprof's UI is in the right direction but not quite as good as Shark's last I checked. Having the X server doing a lot of work on your app's behalf, but in its own process, really hurts profilability though. Rob I tried AQTime a while back. It sucked, I think because it didn’t do call-stack sampling, which is what Shark does and what VTune doesn’t (which is the fundamental reason why VTune sucks, although the UI is horrible too).

Sysprof and oprofile (and dtrace?) do call-stack sampling. Oprofile doesn’t really have UI, sysprof’s UI is in the right direction but not quite as good as Shark’s last I checked. Having the X server doing a lot of work on your app’s behalf, but in its own process, really hurts profilability though.

Rob

]]> By: vladimir http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2265 vladimir Thu, 17 Jul 2008 20:43:02 +0000 http://blog.vlad1.com/?p=61#comment-2265 @fejj: Yeah, Quantify still exists, but the last few times I tried it it either didn't work with more recent versions of Visual C , or just flat out balked at mozilla-sized code. Not a pretty experience. @Miha: I haven't! I remember hearing about AQTime a while ago, but it completely slipped my mind. I've grabbed the eval and will give it a try. They have sane licensing terms as well, so I hope it works out well. @Perry: Yeah, linux has a bunch of great tools here, especially since it's easy to collect data and then massage them for different displays. I've mostly used oprofile and sysprof, but I need to try kcachegrind as well. @Joe: Cool, I'll add GlowCode to the mix. I guess I'll do another followup post once I try out AQTime and GlowCode, and I'll poke at the linux tools as well. @fejj: Yeah, Quantify still exists, but the last few times I tried it it either didn’t work with more recent versions of Visual C , or just flat out balked at mozilla-sized code. Not a pretty experience.

@Miha: I haven’t! I remember hearing about AQTime a while ago, but it completely slipped my mind. I’ve grabbed the eval and will give it a try. They have sane licensing terms as well, so I hope it works out well.

@Perry: Yeah, linux has a bunch of great tools here, especially since it’s easy to collect data and then massage them for different displays. I’ve mostly used oprofile and sysprof, but I need to try kcachegrind as well.

@Joe: Cool, I’ll add GlowCode to the mix.

I guess I’ll do another followup post once I try out AQTime and GlowCode, and I’ll poke at the linux tools as well.

]]> By: Joe http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2263 Joe Thu, 17 Jul 2008 17:09:54 +0000 http://blog.vlad1.com/?p=61#comment-2263 Also, I just remembered one awesome memory/performance profiler that I used/saw used in a previous job: <a href="http://www.glowcode.com" rel="nofollow">GlowCode</a>. It doesn't work on 64-bit code, but I do remember it being pretty awesome. Also, I just remembered one awesome memory/performance profiler that I used/saw used in a previous job: GlowCode. It doesn’t work on 64-bit code, but I do remember it being pretty awesome.

]]> By: Miha http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2262 Miha Thu, 17 Jul 2008 06:50:22 +0000 http://blog.vlad1.com/?p=61#comment-2262 Vlad did you by any chance try AQTime from http://www.automatedqa.com/ ? Vlad did you by any chance try AQTime from http://www.automatedqa.com/ ?

]]> By: asdf http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2260 asdf Thu, 17 Jul 2008 06:28:52 +0000 http://blog.vlad1.com/?p=61#comment-2260 On the other hand, Apple's instrumented profiling is incredibly non-useful, as you will quickly produce traces so big that the analysis tool can't open them. It is buggy as hell in many other ways too. On the other hand, Apple’s instrumented profiling is incredibly non-useful, as you will quickly produce traces so big that the analysis tool can’t open them. It is buggy as hell in many other ways too.

]]> By: Jeffrey Stedfast http://blog.vlad1.com/2008/07/16/why-cant-vtune-be-more-like-shark/#comment-2250 Jeffrey Stedfast Thu, 17 Jul 2008 03:21:46 +0000 http://blog.vlad1.com/?p=61#comment-2250 Heya Vlad, Michael Zucchi and I used to use Rational's Quantify for profiling Evolution back in the day (we had a copy for Solaris) and I /think/ it is available for Windows these days. Of course, it's not free ;-) I also can't really remember how it compares to Shark or VTune other than Quantify instrumented the code, it didn't use intervals iirc. So it probably gets more accurate data overall, but that might not be all that important depending on what you are profiling. The sampling technique used by most tools is probably "good enough". http://www.mozilla.org/unix/quantify.html hehe :-) Anyways, clearly not as quick and simple to get setup as Shark which is one of the down sides to needing to instrument the code :-( Heya Vlad,

Michael Zucchi and I used to use Rational’s Quantify for profiling Evolution back in the day (we had a copy for Solaris) and I /think/ it is available for Windows these days. Of course, it’s not free ;-)

I also can’t really remember how it compares to Shark or VTune other than Quantify instrumented the code, it didn’t use intervals iirc. So it probably gets more accurate data overall, but that might not be all that important depending on what you are profiling. The sampling technique used by most tools is probably “good enough”.

http://www.mozilla.org/unix/quantify.html

hehe :-)

Anyways, clearly not as quick and simple to get setup as Shark which is one of the down sides to needing to instrument the code :-(

]]>