Mailing-List: contact dev-help@archiva.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@archiva.apache.org
Received-SPF: neutral (athena.apache.org: local policy)
Sender: Brett Porter <brett@porterclan.net>
Content-Type: text/plain; charset=iso-8859-1
Mime-Version: 1.0 (Apple Message framework v1251.1)
Subject: Re: Archiva memory settings and stability
From: Brett Porter <brett@apache.org>
In-Reply-To: 
 <CAPoyBqR2Doy65h6a9dSe9vEVZjM-KCVKYXse3qsMztq2wd13+w@mail.gmail.com>
Date: Thu, 5 Jan 2012 22:31:48 +1100
Content-Transfer-Encoding: quoted-printable
Message-Id: <5083D35C-79DD-4B2F-BC49-EB7D34FB85BD@apache.org>
References: <A0F18F9A-6DD6-49D5-B2EE-D8220E7938BC@apache.org>
 <CAPoyBqR2Doy65h6a9dSe9vEVZjM-KCVKYXse3qsMztq2wd13+w@mail.gmail.com>
To: dev@archiva.apache.org


On 05/01/2012, at 7:08 PM, Olivier Lamy wrote:

> 2012/1/5 Brett Porter <brett@apache.org>:
>> Hi,
>>=20
>> In MRM-1549, the -Xms was increased to 512m. This is very high for a =
starting point - would it be better to use the default settings, but let =
it grow to the upper setting? It is quite costly on memory when I run it =
on my local machine.
> Not xmas RAM ? :-)

Hehe, no problem on the main machine (except for the VMs it has going =
too), but it's more useful on the laptop when I'm offline and need to =
conserve RAM and CPU... :)

>>=20
>> A couple of years back I did a lot of profiling to try and get it as =
lean as possible, so that we could run a minimal instance - this is =
something I'd like to do again before 1.4. Between the large number of =
classes to load and other memory use it's becoming impractical to run =
that way again :)
>>=20
> I usually like having -Xms to prevent later slower allocation.
> But as it's something easily configurable: no problem.

Understood - maybe we should also really consider again two =
distributions (minimal proxy & full application).

I guess I can tweak the settings locally as well.

>=20
> BTW removing xmlrpc will decrease number of loaded classes. (and
> moving to a plain html/js will remove struts classes loading too: more
> longer task :-) ).

Right...

>=20
>> Also, I've noticed that the latest milestones are still not very =
stable on vmbuild, restarting once or twice a day. Is anyone else seeing =
similar problems?
>=20
> No I'm not aware of that.
> Any logs or stacktrace or dump available ?

I need to poke at it some more (I think you also have access still?). I =
think it's just a lot of GC but I could be wrong.

- Brett

--
Brett Porter
brett@apache.org
http://brettporter.wordpress.com/
http://au.linkedin.com/in/brettporter
http://twitter.com/brettporter