- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
Does anyone has any experience with running VTune in a virtualized environment ? I have OSX and I am running Windows 8 under VMware Fusion with Intel Parallel Studio XE 2013 installed on it. VMware fusion does have an option so that profiling tools like VTune work on it. Therefore, VTune works on this setup.
I would like to know if those results are reliable. For instance, I have been surprised to find out that from run to run a small microbenchmark code always take the same timing under OSX (with variation under 5% for a timing of 0.5s). Under Fusion/Windows, VTune gives me a variation of time of around 20% from run to run ! But this variation disapears completly with some "real code" that takes around 13s to run. Therefore, I am puzzled with the reliability of VTune in a virtualized environment. Does anyone has some experience with that ?
best regards,
Francois
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
This post just tells us that VMware Fusion supports VTune (more precisely the performance counter in the Intel processors). I already know that. All I want to know is if those results are reliable, and if anyone has experience with it.
Best regards,
Francois
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Sorry I did not pay attention to that part of your post which was related to VMWare Fusion.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I need some more details to try to address your question - although I have no personal experience with collecting on VMware*.
What analysis type were you using?
And, yes, a .5 second run is going to be inherently less accurate because you are reducing the number of samples and sampling is a statistical profiling method. For example, if you use General Exploration and multiplex the events, for a half second run you basically are not collecting enough data per event. For Advanced Hotspots, it may be representative of the code because there are not enough data points.
The other thing to be aware of is the "Duration time estimate" property of the project. The default is "1 to 15 minutes", so you should change it to "under 1 minute" whenever you are profiling an application that runs for less than a minute.
As far as "reliable", hardware-based sampling either works or it doesn't. Since the PMU is exposed by VMware, VTune Amplifier will be able to program the PMU and collect samples. It should be "reliable." The variation you saw is not doubt due to the short run.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Anderson,
Thanks for your feedback. I'll give it a try when I come back to my computer:
Also, there seems to be a bug in the installer that says that VTune won't function properly. As it happens both under VMware/Windows and native Windows, I guess it is a bug in the installer.
Francois
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Is it just a "warning" message? If it detects a VM, it will warn you, but it doesn't stop you from installing or using the product. The software-based collector will work just fine on a VM.

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page