This is the mail archive of the systemtap@sourceware.org mailing list for the systemtap project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Why systemtap loses events and what can be done about it?

Frank Ch. Eigler wrote:
Hi -

On Tue, Sep 04, 2007 at 12:29:34PM -0700, Mike Mason wrote:


snip

Also, why do we only use per cpu variables for aggregations?  Is it
because of memory concerns or something more than that?


It's specifically not for memory concerns.  It's to increase available
parallelism, with different CPUs *adding* data to aggregations
concurrently with <<<.  (In contrast, *reading* data with @count
etc. requires cross-cpu synchronization - modeled by exclusive locks -
to produce meaningful data.)

Sorry, my question was unclear. What I meant was "why don't we use per cpu variables for anything besides aggregations?" I'm sure there's a good explanation (e.g. frequency and cost of cross-cpu synchronization too high). I just want to understand how the design decision was made.

Mike


- FChE

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]