jigsaw startup memory requirements

[Martin Buchholz]

I found that bumping from -Xmx2m to -Xmx2200k makes it pass, so that got
checked in.

Generally, "startup and footprint" remains an ever-important unsolved
problem.  As one part of that, we want to drive down the memory
requirements for Hello World.  I recognize that splitting into multiple
modules allows some of the modules to be omitted, which is a win, but
probably increases the overhead per piece of code included in the runtime,
which is not.  JDK is not very good at managing a large collection of jar
files.

On Fri, Oct 9, 2015 at 2:48 PM, Alan Bateman <Alan.Bateman at oracle.com>
wrote:

> On 09/10/2015 22:15, Martin Buchholz wrote:
>
>> Latest jigsaw b83 works well with the temporary hacks we have in place in
>> jsr166 CVS, but we do see one jtreg test failure.
>>
>> Running a modern jdk with only -Xmx2m is pushing it, but you probably
>> don't
>> want to arbitrarily increase jdk memory requirements just for jigsaw.
>>
>> Should be bump up the -Xmx in this test or wait for y'all-dev to reduce
>> memory footprint?
>>
>> util/concurrent/ConcurrentLinkedQueue/RemoveLeak.java
>>
>> As things currently stand then the heap usage at startup is about 800k
> higher than before when all modules are linked in, much less on a small
> runtime (for embedded) that only has a small number of modules linked in.
> There are few things in flight that should reduce this a bit. We also need
> to re-examination once the compact strings work (JEP 254) goes into JDK 9.
>
> The other thing is the switch to region based heap with G1. Does
> RemoveLeak pass if you specify -XX:+UseParrallelGC?
>
> -Alan
>