pcp
[Top] [All Lists]

Re: Multi-Volume Archive + Live Data Playback for PCP Client Tools

To: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Subject: Re: Multi-Volume Archive + Live Data Playback for PCP Client Tools
From: "Frank Ch. Eigler" <fche@xxxxxxxxxx>
Date: Mon, 3 Nov 2014 21:18:14 -0500
Cc: "'Dave Brolley'" <brolley@xxxxxxxxxx>, "'PCP Mailing List'" <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <54580B6A.1090007@xxxxxxxxxxxxxxxx>
References: <542C21AE.1010504@xxxxxxxxxx> <007e01cfe010$7867f090$6937d1b0$@internode.on.net> <545110DC.2020104@xxxxxxxxxx> <y0mtx2hmwml.fsf@xxxxxxxx> <001d01cff6d7$0e9e4a50$2bdadef0$@internode.on.net> <20141103142643.GA3859@xxxxxxxxxx> <5457F82D.3050307@xxxxxxxxxxxxxxxx> <20141103215816.GC3859@xxxxxxxxxx> <54580B6A.1090007@xxxxxxxxxxxxxxxx>
User-agent: Mutt/1.4.2.2i
Hi, Ken -

> [...]  I was not advocating a fixed mapping [...]

Aha.

> The pmdaCache* routines would allow a dense allocation of more than 4 
> million unique ids and maintain a persistent mapping from the PMNS name 
> that the PMDA chooses and those unique ids.

OK, that could work.

> Perhaps someone could send me the output from:
>       $ pminfo -dm papi
> from a machine where the papi PMDA is running and collecting real data.

(Sent under separate cover.)

> Then restart the PMDA, start a different collection profile and repeat 
> the command above.

It won't change just based on that.  But a different version of
PAPI/libpfm, a CPU/kernel change, can easily generate a different
enumeration of available events, thus a different and/or reordered
PMNS name sets.

- FChE

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