Flag in JDK 7uXX to allow visibility into code cache usage?
ysr1729 at gmail.com
Mon Nov 4 14:19:22 PST 2013
In 8, I see that there is a product flag called
"PrintCodeCacheOnCompilation" which reports the occupancy etc. each time a
method is compiled.
(Perhaps that's a bit too much/frequent, I don't know.) There's also a
develop flag "PrintMethodFlushing" that (in 8) reports this at the end of
an nmethod sweep cycle.
(Would be great if this were printed in product mode, not just develop.)
It would be great if the latter were also reported in product mode and the
reporting back-ported to 7uXX.
On Mon, Nov 4, 2013 at 1:25 PM, Kirk Pepperdine <kirk at kodewerk.com> wrote:
> Hi Ramki,
> My little VisualVM plugin MemoryPoolViewer was motivated by a need to see
> the occupancy and size of the code cache. I think it would be helpful if
> the compilation threads reported on pool maintenance as the GC threads do.
> On 2013-11-04, at 6:17 PM, Srinivas Ramakrishna <ysr1729 at gmail.com> wrote:
> > We recently had a situation when code cache flushing, when the code
> cache was nearly full, caused a large drop
> > in performance. It would be nice if there were a product flag (also in
> 7uXX) by which one could get
> > a read of the code cache size on a periodic basis. It might be great,
> for instance, if this value were
> > printed, say at each GC.
> > I understand that there is an Mbean that can be read to determine this
> value, but I was looking for
> > something like a JVM flag that allows this to be printed to a log file
> on a periodic basis, say at each GC (or even eacn
> > major gc cycle). In 7uXX at least there seem to be far too few means to
> track the usage of the code
> > cache.
> > thanks for anything that can be done in this area, wrt visibility (in
> 7uXX :-)
> > -- ramki
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the hotspot-compiler-dev