Serial old to young root scan

Srinivas Ramakrishna ysr1729 at gmail.com
Mon Nov 14 21:20:22 UTC 2011


Hi Lokesh --

AFAIR, it's just a historical artifact.  (Jon, Peter or John will probably
know for sure.)
 If the perm gen card-scanning task is queued first, its longer latency
will usually be hidden behind the parallel old gen scanning tasks. My guess
is that with the sizes of
perm gen that it was tested with, this didn't matter, and may be it now
does with the sizes you are
looking at? This may become moot with the new perm-gen-less meta-data
space, where perhaps
a different parallel (?)  mechanism will be employed for scanning these
"roots into young".

Have you obtained data with +PrintGCTaskTimes (or such) that indicates that
this is
an issue?

-- ramki

On Mon, Nov 14, 2011 at 6:33 AM, Lokesh Gidra <lokesh.gidra at yahoo.com>wrote:

> Hello,
>
> Is there any specific reason why the permanent generation old to young
> roots are scanned serially, unlike old generation old to young roots, in
> Parallel Scavenge GC?
>
>
> Kind Regards,
> Lokesh
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20111114/96389f2f/attachment.htm>


More information about the hotspot-gc-dev mailing list