How Intrusive Is VTune Driver?

[C_Michael_Martell]

It appears that the VTune driver needs to not only be installed,but running before any collector options can be changed for remote data collection. This seems like an oversight in rdc.

So my question is, given that the driver needs to be running, how intrusive is it? Since we're trying to use rdc on a portal machine (i.e., strictly acting as secure access to target machines),the system administrators are concerned that running the VTune driver on this machine will be too costly in terms of machine resources.

Thanks, Mike.

[jeffrey-gallagher]

Very interesting question mmartell,

It's been our experience in the VTune group that most users don't write development code in a production environment with other users. Most are in a fairly controlled development systemenvironment (for a variety of reasons).

That said, I note that on my lab system, I have exactly 21 kernel modules loaded up at any given time, all but one of which were already on the system by virtue of my default install:

# lsmod

VTune is one of them, of course. In terms of expense, we consider it extremely minimal although I've never tried to measure it in relative terms with other kernel modules.

The module works with the /dev/vtune* files that are also installed, and is in play only during sampling sessions (callgraph does not use the sampling drivers).

It seems much more likely that the code you're running would be have a negative impact on system resources, affecting other users, than this small driver. That might be enough reason not to sample on a production system all by itself.

Also please note, depending on what sorts of sampling experiments you set up, VTune can collect huge amounts of data in a very short time. You'd want to use a system with lots of available space to insure no unpleasantness with regard to user quotas, or almost-full filesystems filling up.

cheers

jdg