Analyzers
Talk to fellow users of Intel Analyzer tools (Intel VTune™ Profiler, Intel Advisor)
4974 Discussions

Port collision / does Linux VTune use any well-known ports?

Deleted_U_Intel
Employee
630 Views

Hi,

I'm running an evaluation copy of VTune to see if it
works with our server product that runs on Redhat 7.1.
I'm running VTune 1.1.

Its giving me mixed results. I'm finding when I have
a client program just connect to our server, VTune
seems to think our server has terminated (and it hasn't).

VTune gives the message:

Caught signal 2, SIGINT (interrupt)
Fri Nov 14 08:22:03 2003 mexia (Run 0) All sampling data collected during this activity has been deleted because the collection was cancelled.
The Activity has finished running.

But in fact, our server is still running.
This looks like a spurious error to me.

I know our server's portmapper is using ports
2114, 2120 and 2121. I just guessing, but I'm
suspecting that since this error seems to occur
when a client program connects, perhaps there's
some port collision confusing VTune.

Anyone have any ideas about this? VTune and socket
ports?


Thanks!


John Roberts
john_roberts@credence.com

0 Kudos
5 Replies
jeffrey-gallagher
630 Views
Hey John Roberts,

Maybe the best thing here is to open a premier case asap and report this directly to our support team. That said:

1) Note that there is a 20-second default sampling duration for sampling sessions. Try running your session with a longer duration, -d 5000 (for example) and see if you get better results.

NOTE: that 20-second session default is going away when 2.0 comes out, if a user indicates a -app at the command line, anyway.

2) If you still have trouble, try running the ActivityController to gracefully shut down the activity, which will actually generate results you can view.

If you haven't used it before:

$ man ActivityController

note upper case A and C.


If neither of these ideas assists you, get that premier case opened and let's let the support dudes chew this one up for breakfast.

:)

cheers
jdg

0 Kudos
jeffrey-gallagher
630 Views
John_Roberts,

Ah, for beta, we turned calibration ON by default. That means your app runs twice when you sample: once for VTune to get a feel for what your app is doing, and the second run is for the actual activity you've asked for itself.

You'll need a -o to turn off calibration, like this example:

$ vtl activity -c sampling -o "-cal no" run

(you'd still keep your -d in there, too, of course)


NOTE: the released version of this software will have calibration turned OFF by default, so users won't be bugged by this very issue soon.

cheers

jdg

0 Kudos
jeffrey-gallagher
630 Views
Good detective work, john_roberts. Glad you got it going!
cheers
jdg
0 Kudos
Aaron_L_Intel
Employee
630 Views

Hi John,

It makes sense that you would need to use -d after -o instead of in between -c and -o, because -o must follow the collector specification, unlike other options for the activity command. It is possible to specify multiple collectors on the command line, and because of this, -o "[options]" must follow -c [collector name]. -d, along with the other activitycommand options, can go anywhere, as long as they don't fall between -c [collector name] and -o "[options]".

The output of vtl -help activityattempts to get this message across:

Usage:
activity [] Create Activity with optional name .
[-d | -duration <#>] Use <#> as duration, in seconds, for Activity.
[-master ] Use as master collector for multi-collector
Activity.
[-start-paused] Start Activity with data collection paused.
{-c | -collector [,]}
Add collector with optional name .
[-o | -options ]
Specify collector options in . Embed the
options within double quotes.
[-of | -options-from-file ]
Specify collector options from file .
[-app | -app-to-launch [,[,

]]]
Specify an application to launch with path ,
optional command-line arguments , and
optional working directory .
[-moi | -module-of-interest ]
Specify a module of interest with path .


AaronL

0 Kudos
jeffrey-gallagher
630 Views
Well said, John.
The beta of the next release of this software is due out Q2 2004. Like the objects in my car's rear view mirror, that's a lot closer than it sounds. Er, appears.
It will have a GUI, which will assist with some of these interface issues. Stay tuned here for more announcements as to when that software can be tried, but I think you'll agree it will be a step up from the current confusion you're feeling.
Thanks for reporting in. This kind of feedback is PRECISELY the reason this forum exists. Keep up the great work.
cheers
jdg
0 Kudos
Reply