pcp
[Top] [All Lists]

Re: PAPI pmda Note

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: PAPI pmda Note
From: fche@xxxxxxxxxx (Frank Ch. Eigler)
Date: Wed, 17 Sep 2014 16:51:50 -0400
Cc: Lukas Berk <lberk@xxxxxxxxxx>, pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <1161038122.50704119.1410924350821.JavaMail.zimbra@xxxxxxxxxx> (Nathan Scott's message of "Tue, 16 Sep 2014 23:25:50 -0400 (EDT)")
References: <20140905200756.GA31071@xxxxxxxxxx> <214253475.50237185.1410857725513.JavaMail.zimbra@xxxxxxxxxx> <87fvfr5ipk.fsf@xxxxxxxxxx> <1161038122.50704119.1410924350821.JavaMail.zimbra@xxxxxxxxxx>
User-agent: Gnus/5.1008 (Gnus v5.10.8) Emacs/21.4 (gnu/linux)
Nathan Scott <nathans@xxxxxxxxxx> writes:

> [...]  OOC, is it possible to ask PAPI what hardware counters are
> enabled?  (IOW, without relying on the state kept in-memory, in the
> daemon, as a result of changes it made itself, but rather by asking
> the hardware/kernel?).

No - the kernel keeps that info to itself, and multiplexes amongst
applications and its own internal usage.


> If so, that would be great to use as the value for the control "enable"
> metric, 

(I've nudged lberk toward using the ".status" read-only metric as a
better-colored toolshed to describe current state.)

> instead of simply exporting whatever state changes we've made in the
> PMDA (and assuming everything was disabled when it started, which
> may not have been the case). [...]

(I've nudged lberk toward not making ".enable"/".disable" readable at
all, especially if they are to return merely the most-recent deltas
rather than the cumulative situation.)


- FChE

<Prev in Thread] Current Thread [Next in Thread>