[KLUG Members] Load distribution on SMP system

[Adam Tauno Williams]

>> But your load average is only 1;  it looks to me, that while you may
>> have multiple Java threads, they have nothing to do and are not
>> dinging
>> the schedular.  Either the work just isn't there to be done or there
>> is
>> some type of configuration problem in the *application*.  If multiple
> t>hreads were held in a pending state then load average would climb
> above 1.x
> Yep, I figured it out later. While the other threads were idle, one
> was doing garbage collection - the offending thread above.
> I'm surprised that GC is not distributed among the available CPUs,

Since they are all in the same VM (threads) there is only one page-map, which
the GC must lock to do collection, so it isn't a process that can be 
chunked up
as the pretty much the whole GC process exists in a critcal section stanza.

> though - maybe my understanding of GC is faulty. I did read this:
> "If not managed, garbage collection is one of the biggest bottlenecks
> for an application, especially when running on symmetric
> multiprocessing (SMP) server machines."
> (http://publib.boulder.ibm.com/infocenter/wasinfo/v5r0/index.jsp?topic
> =/com.ibm.wasee.doc/info/ee/ae/rprf_javamemory.html)