pcp
[Top] [All Lists]

Re: Java JNI PMAPI

To: steve@xxxxxxxxxxxxx
Subject: Re: Java JNI PMAPI
From: Mike Werner <mtw@xxxxxxxxxxxxxx>
Date: Fri, 8 Apr 2005 10:23:42 -0700
Cc: pcp@xxxxxxxxxxx
In-reply-to: <f0752656050406055947480b88@mail.gmail.com>
References: <20050331132108.A7227@corp.theblr.com> <Pine.LNX.4.61.0504011206010.11109@kenj-pc.melbourne.sgi.com> <20050401085543.A10490@corp.theblr.com> <f0752656050401172734a6c364@mail.gmail.com> <20050403102558.A18789@corp.theblr.com> <f0752656050404044943a28eb7@mail.gmail.com> <20050406000707.B26376@corp.theblr.com> <f0752656050406055947480b88@mail.gmail.com>
Sender: pcp-bounce@xxxxxxxxxxx
Hi Steve,

I'm glad to hear I was mistaken.  I'd love to have a look at the
JNI/C++ code that you have.  Will you send me a tar.gz ? 

The Tomcat-based system you are describing reminds me of Sun's
Management Center product.  Users/clients connect to a central
server that, in turn, connects to the several target machines.
If I remember correctly, the SunMC server is also Java-based.
I like that scheme for it's security features.  For performance,
dependability, and flexibility I prefer a decentralized scheme,
a la PCP.

Would SGI's jmcd daemon suit your purpose?

- mtw

   Stephen Przepiora wrote on Wed, Apr 06, 2005 at 08:59:27AM -0400

   You got me wrong. I allways give short answers :) You can have total control 
   over it as the basic architecture is flawed for me. I _NEED_ to use it in a 
   java multithreaded environment (tomcat). Synchronization would not work as 
   it would slow down collection of the metrics to much. I was going to move 
   into a deamon with a simple protocol much like what Ken talked about.
   
   The JNI interface really is simple. It is a basically a wrapper around a 
   bunch of C++ classes I wrote to use the pmapi. I think that is much more 
   interesting than the JNI interface.
   
   Moving forward I was planning on dumping the JNI interface, and using the 
   C++ classes to write a daemon that will do the PCP work. Then write some 
   java classes that would connect to the daemon and get the metrics.
   
   Steve
   
   On Apr 6, 2005 3:07 AM, Mike Werner <mtw@xxxxxxxxxxxxxx> wrote:
   > 
   > Hi Steve,
   > 
   > Ok. The rest seems pretty straightforward from there.
   > From your short answer I'm guessing that you're not
   > interested in having any outside involvement. Feel
   > free to drop me a line if you want to chat or tell
   > any more. Happy coding.
   > 
   > Cheers,
   > 
   > - mtw
   > 
   > Stephen Przepiora wrote on Mon, Apr 04, 2005 at 06:49:49AM -0500
   > 
   > It will fetch metrics.
   > 
   > On Apr 3, 2005 12:25 PM, Mike Werner <mtw@xxxxxxxxxxxxxx> wrote:
   > >
   > > Hi Steve,
   > >
   > > Regarding Java, I have just recently begun exploring the
   > > idea of a JNI implementation. You seem much further along.
   > >
   > > Are you planning to release it?
   > > Are you interested in collaborating?
   > > How complete is your wrapping of the PMAPI?
   > >
   > > - mtw
   > >
   > > Stephen Przepiora wrote on Fri, Apr 01, 2005 at 08:27:32PM -0500
   > >
   > > Have you tried JNI? I have the beginnings of a JNI implementation, but
   > > stopped because I needed to use it in a threaded enviroment. I am
   > > currently
   > > reworking how the application will run.
   > >
   > > Steve
   > >
   > > On Apr 1, 2005 11:55 AM, Mike Werner <mtw@xxxxxxxxxxxxxx> wrote:
   > > >
   > > > Hi Ken & Mark
   > > >
   > > > Thanks for the helpful info.
   > > >
   > > > Ken - as to why: I'm exploring some hairbrained ideas
   > > > for using pcp data without C, e.g. Java. Do you know
   > > > of any ports for alternate languages?
   > > >
   > > > - mtw
   > > >
   > > > kenmcd@xxxxxxxxxxxxxxxxx wrote on Fri, Apr 01, 2005 at 12:15:11PM 
   > +1000
   > > >
   > > > On Thu, 31 Mar 2005, Mike Werner wrote:
   > > >
   > > > > PCP Team,
   > > > >
   > > > > Is there a specification (ASN.1 or otherwise) of the
   > > > > PMCD-to-client protocol, which I might obtain?
   > > >
   > > > I can assure you it is _not_ ASN.1 ... PCP is for analyzing 
   > performance
   > > > problems, not creating them ... 8^)>
   > > >
   > > > The message protocol and format is not a secret
   > > >
   > > > pminfo -D pdu -v
   > > >
   > > > will dump out the messages flowing in both directions between the
   > > > client and pmcd.
   > > >
   > > > We don't have any documentation beyond that ... but more to the point
   > > > I'm kinda curious as to why you're interested in this, as the only
   > > > sensible way to use the infrastructure is via libpcp and there is
   > > > detailed documentation available on that API.
   > > >
   > > > Roughly, each libpcp call maps onto 1 sent message and 1 received
   > > message.
   > > >
   > > >
   > >
   > >
   > 
   >

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