Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Kristian Rosenvold
2012/12/12 Igor Fedorenko : > > > On 2012-12-12 11:35 AM, Kristian Rosenvold wrote: >> >> Does that mean m2e ditches the container every now and then? >> > > No. m2e keeps the same container but injects its own cache > implementations that allows purging project-specific cache entries > whenever wo

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Christian Schulte
Am 12/12/12 08:23, schrieb Milos Kleint: > how much memory will be freed by the soft reference? if it's not a big > chunk, most likely not worth it, soft references are released only > when your VM is really, really in trouble. by the time it gets > released, you've been slowed down by repeatedly h

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Jason van Zyl
On Dec 12, 2012, at 12:09 PM, Igor Fedorenko wrote: > > > On 2012-12-12 11:35 AM, Kristian Rosenvold wrote: >> Does that mean m2e ditches the container every now and then? >> > > No. m2e keeps the same container but injects its own cache > implementations that allows purging project-specific

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Igor Fedorenko
On 2012-12-12 11:35 AM, Kristian Rosenvold wrote: Does that mean m2e ditches the container every now and then? No. m2e keeps the same container but injects its own cache implementations that allows purging project-specific cache entries whenever workspace project is re-read or removed from w

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Jason van Zyl
In embedded mode the ITs were failing without the change. I can't remember what you were looking at but what problem is it causing? jvz On 2012-12-12, at 11:35 AM, Kristian Rosenvold wrote: > Does that mean m2e ditches the container every now and then? > > In that case the whole unloading j

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Kristian Rosenvold
Does that mean m2e ditches the container every now and then? In that case the whole unloading jason implemented can be reverted...? K Den 12. des. 2012 kl. 16:44 skrev Igor Fedorenko : > For tests, I think the easiest is to check number of realms at the end > of each test and drop plexus contai

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Igor Fedorenko
For tests, I think the easiest is to check number of realms at the end of each test and drop plexus container if it grew over certain number of realms. Pick the number large enough to fit in 128M of permgen and I think this will provide good tradeoff between performance and memory usage. This does

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-12 Thread Kristian Rosenvold
2012/12/12 Milos Kleint : > how much memory will be freed by the soft reference? if it's not a big > chunk, most likely not worth it, soft references are released only > when your VM is really, really in trouble. by the time it gets > released, you've been slowed down by repeatedly hitting the ceil

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-11 Thread Milos Kleint
how much memory will be freed by the soft reference? if it's not a big chunk, most likely not worth it, soft references are released only when your VM is really, really in trouble. by the time it gets released, you've been slowed down by repeatedly hitting the ceiling of your memory and CGed freque

Re: A very interesting performance regression in 3.1 embedded mode

2012-12-09 Thread Christian Schulte
Am 12/09/12 22:58, schrieb Kristian Rosenvold: > > Anyone else have any ideas about eviction strategies ? > Without having looked at the code. GC driven by using soft references. -- Christian - To unsubscribe, e-mail: dev-un

A very interesting performance regression in 3.1 embedded mode

2012-12-09 Thread Kristian Rosenvold
Now that 3.1 works with embedded mode again I have found an /interesting/ performance regression; In this commit https://git-wip-us.apache.org/repos/asf?p=maven.git;a=commit;h=451c43152b8939a699dfa5db4fc4ca8676182462 the plugin realm cache is purged after each embedded build. While this obviously