[ale] Load?

Christopher Fowler cfowler at outpostsentinel.com
Sat May 21 16:20:26 EDT 2005


I've seen this happen with threads.  I've got a PIV system that runs
tomcat.  We have a java thread that can go haywire after a remote
connection terminates itself uncleanly.  Tomcat will not show yup in the
top of top.  Top shows nothing that seems to be eating cpu.  It took me
a day and strace to locate that problems causing our load averages to be
so high.  I attached to each process to see if they were busy in system
calls.  The java program would read 0 bytes on a fil handle and just
keep doing it forever.  It could execute as many read() calls as it
possibly could and it would eat away at the CPU.  After a few of these
threds the system would go to a crawl but top still show nothing
abnormal other than high load numbers.
  
On Fri, 2005-05-20 at 19:57, David Corbin wrote:
> On Thursday 19 May 2005 10:33 pm, Michael B. Trausch wrote:
> > David Corbin wrote:
> > > top - 20:43:03 up 18 days,  7:57,  2 users,  load average: 4.00, 4.00,
> > > 4.00 Tasks:  65 total,   1 running,  64 sleeping,   0 stopped,   0 zombie
> > > Cpu(s):   0.7% user,   0.0% system,   0.0% nice,  99.3% idle
> > >
> > > Here are the top few lines from a top.  The system is substantially idle,
> > > yet the load reads a consistent '4'.
> > >
> > > How do I track the source of this behavior down?
> >
> > That's interesting...  Is there a unique situation on your box that is
> > likely to trigger a bug in the kernel, which reports system load?  
> 
> Not that I'm aware of :)
> 
> > A 
> > system load of 4.00 would be consistant with 4 processes trying to get
> > 100% of the CPU time, and consequently, is typically enough to keep a 4
> > CPU system pretty busy.
> 
> It's not a 4CPU system, and the CPUs are shown as substanitally idle.
> 
> >
> > What kernel version are you running?
> 
> It's a Debian system running 2.4.23.
> 
> >
> > 	- Mike
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://www.ale.org/mailman/listinfo/ale



More information about the Ale mailing list