Re: Lots of dbsnmp and jre processes, memory hug

On Wed, 29 Oct 2003 22:39:00 +0100, Rick Denoire wrote:

> RedHat Advanced Server 2.1 / Dell PE2650 (2 CPUs, Intel Xeons), Oracle
> 9.2.0.4
>
> I have been trying to set up the Intelligent Agent and the Oracle
> Management Server (OMS) in order to use the Enterprise Manager and
> lots of utilities in the Diagnostic Pack.
>
> Thanks to Oracle Support, now I can connect to the OMS using the OEM
> (in Servers with more than one network connection, it won't work out
> of the box).
>
> But after starting the agent (which is the data gatherer too), more
> than one dozen of dbsnmp processes start. And if the OMS is startet, a
> couple of dozens of jre processes appear. These jre processes consume
> about 1.5 GB of RAM (!) no matter if there is a connection from the
> OEM or not.
>
> Two days after connecting to the OMS (the OEM kept open and connected
> all the time), all four instances running on the server got stuck. No
> connection possible, nothing. My first suspicion was right and I just
> closed the OEM, all instances came back to life immediately.
>
> This seems to be strange, but what is really weird, is the kind of
> messages that I found in the alert logs. I can't even recall them (not
> in office now) and I think that I will have to open a TAR again.
>
> Has anyone here had similar experiences running OMS/OEM? Are these
> Linux specific problems?
>
> Bye
> Rick Denoire

Actually, top is deceiving you. Those are not "jre processes", those are "jre threads". Threads are lightweight processes with kernel context only which map user area of given process that executes them. Your jre is multithreading and your tool is mistakenly reporting all those lightweight processes as full-fledged processes. That, precisely, is the rason why you no longer have gtop on RH. There is some inane "monitor" thingy but my impression is that nobody really knows what to do with those threads. There are several differences between LWP and real processes.

1. LWP doesn't have a user context. It gets one when it gets assigned. It's user area (.u area) is the .u area of the process using it.
2. All threads sharing the same .u area are scheduled simultaneously. They share the same OS priority and they are mad runnable all together. That means that the user decides which thread runs and which one waits. That, of course, is done via threading library.

As LWP's take entries from the process table, top sees them as processes and that is why you're getting confused.

-- 
None of us is as dumb as all of us.
(http://www.despair.com/meetings.html)