indy and tailc

Ismael Juma mlists at juma.me.uk
Wed Jan 27 09:32:13 PST 2010


On Wed, Jan 27, 2010 at 9:42 AM, Y. Srinivas Ramakrishna
<Y.S.Ramakrishna at sun.com> wrote:
> Patrick Wright wrote:
> ...
>>
>> Glad to hear that perm gen will eventually go away.
>
> Could you elaborate a little on why it would be nice if the
> perm gen went away? Where would you like its current contents
> (which ones, if any, specifically?) to be located and why?

Part of the problem with the perm gen is the low value of MaxPermSize
by default. This was set long ago, but it's not suitable for large
applications today, particularly 64-bit applications. A good example
of this is Eclipse that reaches the maximum quite easily. To make
matters worse, one doesn't always get an OOM, sometimes the app just
hangs with high CPU usage. Finally, there is no standard way to change
the value and some JVMs quit when they receive an unknown -XX option.

As a result, Eclipse went for a long time without setting this so
almost every user would run into the problem and have to google for it
(sometimes suffering for quite a bit before doing so). Also, some
users get confused on why they're getting a OOM if their heap (-Xmx)
is huge. None of this helps less experienced users.

The other side of the coin is the issue with redeployments where
people would get a OOM after X redeployments because something was
holding on to something that would prevent the classloader from being
collected. That also gave the perm gen a bad rep (note that I am not
claiming that it is deserved or undeserved, just stating what happened
:)).

Best,
Ismael


More information about the mlvm-dev mailing list