[Advanced-java] Multithreading problem.

[rana@xxxxxxxxxxxxxx (Hemendra Rana)]

It might not be round robin. I am not sure how scheduling works so while
programming I always assume that some random thread is going to be scheduled.

Rogelio Perez wrote:

> Thanks Hemendra, all of my threads are waiting to read in a socket, it could
> be the I/O operation
> you say.
> But in my machine I could spawn more than 999 threads, I don´t know the
> Solaris lwp limits?
>
> Also I´ve seen that sometimes one thread recently created can be scheduled
> before others waiting to be scheduled, Any idea?. Sometimes one thread waits
> about 5 seconds because  new threads are been continously started?. This
> happen when  all of them are waiting to entry in an object monitor. (I don´t
> understand why, --> round roubin?)
>
> Thanks in advance.
>
> ----- Original Message -----
>
> From: "Hemendra Rana" <rana@xxxxxxxxxxxxxx>
> To: "Rogelio Perez" <rogeliop@xxxxxxxx>
> Cc: "Peter Humphries" <Peter.Humphries@xxxxxxxx>;
> <advanced-java@xxxxxxxxxxxxxxxxxxxxxx>; "Hemendra Rana"
> <rana@xxxxxxxxxxxxxx>
> Sent: Wednesday, February 28, 2001 4:12 PM
> Subject: Re: [Advanced-java] Multithreading problem.
>
> You are right in the way that each LWP executes exactly one thread at any
> given
> moment and it would not perform any timeslicing among the eligible threads
> until
> ofcourse it finds a higher priority thread. In native threads model, the JVM
> thread scheduling is more or less controlled by LWP scheduling done by the
> OS.
> So when the LWP gets scheduled to run, it picks up the highest priority JVM
> thread from the pool of JVM threads waiting to be scheduled. The LWP
> eventually
> loses its timeslice and another LWP gets scheduled by the OS. The LWP that
> got
> scheduled now, would look for the highest priority JVM thread waiting to be
> scheduled and would run that thread. When an LWP is scheduled out by the OS
> to
> make way for another LWP, it deposits the JVM thread that was associated
> with it
> into the pool of thread that is waiting to be scheduled.
>
> In case when there are more then one threads of equal priority, waiting in
> the
> pool of JVM threads to be scheduled, the thread that needs to be associated
> to
> the LWP may be chosen in round robin fashion.
>
> The exceptional case occurs when the JVM thread blocks for an I/O. In this
> particular case, the LWP gets stuck with the JVM thread, until the I/O is
> complete. If all the LWPs in the system are stuck for I/O and there are
> threads
> in the pool of JVM threads to be scheduled, the OS assigns a new LWP to the
> JVM
> that then works just like any other LWP.
>
> Chapter 6 of Java Threads by Scott Oaks & Henry Wong (2nd Edition),
> describes
> how java native threads model works on Solaris. So a JVM might create N
> threads,
> but that does not necessarily mean that there are going to be N LWPs. There
> will
> be M LWPs in the system where M <= N. If all the LWPs are not involved in
> I/O
> then M < N and M = N only when all the JVM threads are are blocked for I/O.
>
> Please correct me if I am wrong.
>
> Hemendra
>
> Rogelio Perez wrote:
>
> > but when I execute ps -eLc I could see that 6000 lwp appear in my java
> > process.???
> >
> > Also I´m running my program with java -xss 60k, this option causes a minor
> > memory consume per thread.
> >
> > ----- Original Message -----
> > From: "Hemendra Rana" <rana@xxxxxxxxxxxxxx>
> > To: "Peter Humphries" <Peter.Humphries@xxxxxxxx>
> > Cc: <advanced-java@xxxxxxxxxxxxxxxxxxxxxx>
> > Sent: Tuesday, February 27, 2001 7:56 PM
> > Subject: Re: [Advanced-java] Multithreading problem.
> >
> > > It is an incorrect assumption.
> > >
> > > Peter Humphries wrote:
> > >
> > > > I assumed that with the "-native" option there would be a native
> thread
> > (or
> > > > light-weight process in Solaris speak) per Java thread. Is that an
> > incorrect
> > > > assumption?
> > > >
> > > > Peter H
> > > > Jyra Research Ltd
> > > >
> > > > ----- Original Message -----
> > > > From: "Hemendra Rana" <rana@xxxxxxxxxxxxxx>
> > > > To: "Peter Humphries" <Peter.Humphries@xxxxxxxx>
> > > > Cc: <advanced-java@xxxxxxxxxxxxxxxxxxxxxx>
> > > > Sent: 27 February 2001 18:37
> > > > Subject: Re: [Advanced-java] Multithreading problem.
> > > >
> > > > > The stack address space that you are talking about is the LWP stack
> I
> > > > suppose
> > > > > and not the JVM thread. He has 6000 threads doesn't mean he has 6000
> > lwps.
> > > > >
> > > > > Peter Humphries wrote:
> > > > >
> > > > > > I'm astonished you can get up to 6000 threads at all. I would have
> > > > thought
> > > > > > the stack address space requirements would have blown the (2^32?)
> > limit
> > > > well
> > > > > > before then. Did you have to tweak any settings to get that far?
> > > > > >
> > > > > > Peter H
> > > > > > Jyra Research Ltd
> > > > > >
> > > > > > > -----Original Message-----
> > > > > > > From: Rogelio Perez <rogeliop@xxxxxxxx>
> > > > > > > To: advanced-java@xxxxxxxxxxxxxxxxxxxxxx
> > > > > > > <advanced-java@xxxxxxxxxxxxxxxxxxxxxx>
> > > > > > > Date: Tuesday, February 27, 2001 12:15 PM
> > > > > > > Subject: [Advanced-java] Multithreading problem.
> > > > > > >
> > > > > > >
> > > > > > > Hi,
> > > > > > > I´m working with Java 1.2.2.006 in Solaris 2.7, Sparc Ultra 250,
> 1
> > Gb
> > > > of
> > > > > > > memory.
> > > > > > > My program spawns over 6000 threads. Then I observe that  new
> > threads
> > > > over
> > > > > > > this number are created more slowly.
> > > > > > >
> > > > > > > I run the program with this command line "java - native
> myprogram"
> > > > > > > (native threads)
> > > > > > >
> > > > > > > Does anybody know any limitation in the JVM about number of
> > threads
> > > > that
> > > > > > can
> > > > > > > be executed simoultaneously without performance problems.
> > > > > > >
> > > > > > > Any idea.
> > > > > > > Thanks in advance.
> > > >
> > > > _______________________________________________
> > > > Advanced-java mailing list
> > > > Advanced-java@xxxxxxxxxxxxxxxxxxxxxx
> > > > http://lists.xcf.berkeley.edu/mailman/listinfo/advanced-java
> > >
> > > _______________________________________________
> > > Advanced-java mailing list
> > > Advanced-java@xxxxxxxxxxxxxxxxxxxxxx
> > > http://lists.xcf.berkeley.edu/mailman/listinfo/advanced-java