very slow perm gen verification
John.Coomes at oracle.com
Thu May 12 13:39:26 PDT 2011
Y. Srinivas Ramakrishna (y.s.ramakrishna at oracle.com) wrote:
> Correction. My fix applied to any non-young generation.
> I should has said "non-young gen" instead if "perm gen"
> below. Do you see the issue only with the perm gen?
Yes, I've only seen it with the perm gen. So far I've only tried par
scavenge; I'll try some other GCs.
> On 5/12/2011 11:49 AM, Y. Srinivas Ramakrishna wrote:
> > With the framework collectors there was a badness that
> > would cause O(n^2) in the size of the perm gen in the phase
> > that checks that clean cards do not have cross-generational
> > pointers. I fixed it a few weeks ago so it is now O(n) -- i can dig up the
> > CR if you can't find it. Are you seeing this with a specific
> > collector or all?
> > -- ramki
> > On 5/12/2011 10:33 AM, John Coomes wrote:
> >> John Coomes (John.Coomes at oracle.com) wrote:
> >>> When running some nsk tests, I saw several timeouts which appear to be
> >>> due to very slow verification of the perm gen (i.e., 10+ minutes to
> >>> verify a 64MB perm gen (par scavenge)), which is done as the jvm is
> >>> exiting. Happens frequently, but not every time; and w/my changes as
> >>> well as w/jdk7-b141.
> >>> I vaguely recall some discussion of this a few months ago, but can't
> >>> find any record of it :-(. Sound familiar to anyone?
> >> In case it's not obvious to some, the verification is done only in
> >> non-product builds.
> >> -John
More information about the hotspot-gc-dev