Fwd: [Vm-dev] unused-method removal vs. Spur compaction

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Fwd: [Vm-dev] unused-method removal vs. Spur compaction

Ben Coman
Forwarded from squeak & vm lists in case its useful to others...

On Sun, Sep 18, 2016 at 5:11 AM, Craig Latta <[hidden email]> wrote:

>
>
> Hi--
>
>      I've got the mass unused-method removal from Spoon working in Spur
> in Squeak 5.1. But now I have a juicy 12-megabyte chunk of free space in
> the middle of the 35-megabyte image that won't go away. There's just one
> Spur segment (I was expecting several?), with no pinned objects.
>
>      Is there a known Squeak image with multiple segments? Is there some
> way to split a segment? Am I just supposed to play with the grow
> overhead and shrink threshold in the VM simulator, until the compaction
> I want is triggered?
>
>      thanks!
> Craig Latta



---------- Forwarded message ----------
From: tim Rowledge <[hidden email]>
Date: Mon, Sep 19, 2016 at 1:17 AM
Subject: Re: [Vm-dev] unused-method removal vs. Spur compaction
To: Squeak Virtual Machine Development Discussion
<[hidden email]>



Eliot mentioned using the vmaker sim to load the image and then save
it after removing the empty space. Ah, found it -

> Build a VMMaker image with image/buildspurtrunkvmmaker.sh and you can now use Spur32BitPreen to rewrite the image to be more compact.  We can use this as part of the release process until Clément and I have fixed Spur compaction.  e.g.
>
>       Spur32BitPreen new preenImage: '../oscogvm/image/trunk50'
>       Looking for module  ... loaded...computing accessor depths...done
>       Looking for module  ... loaded...computing accessor depths...done::::.............................................done.
>       old heap size: 41,897,472       initial new heap size: 26,818,472
>       change: -35.99%
>       final new heap size: 26,818,472 change: -35.99%
>
>       Done!
>
tim

Reply | Threaded
Open this post in threaded view
|

Re: Fwd: [Vm-dev] unused-method removal vs. Spur compaction

Clément Béra
I am not sure this is the right mailing list.

Spur32BitPreen could be used as part of the Pharo release process as long as the compactor is not working well. This way released images can be more compact. If the Pharo folks wants to run Spur32BitPreen from Pharo to compact the image, the time spent to make things work may not be worth it as the mid-term solution is a new compactor.

On Mon, Sep 19, 2016 at 1:42 AM, Ben Coman <[hidden email]> wrote:
Forwarded from squeak & vm lists in case its useful to others...

On Sun, Sep 18, 2016 at 5:11 AM, Craig Latta <[hidden email]> wrote:
>
>
> Hi--
>
>      I've got the mass unused-method removal from Spoon working in Spur
> in Squeak 5.1. But now I have a juicy 12-megabyte chunk of free space in
> the middle of the 35-megabyte image that won't go away. There's just one
> Spur segment (I was expecting several?), with no pinned objects.
>
>      Is there a known Squeak image with multiple segments? Is there some
> way to split a segment? Am I just supposed to play with the grow
> overhead and shrink threshold in the VM simulator, until the compaction
> I want is triggered?
>
>      thanks!
> Craig Latta



---------- Forwarded message ----------
From: tim Rowledge <[hidden email]>
Date: Mon, Sep 19, 2016 at 1:17 AM
Subject: Re: [Vm-dev] unused-method removal vs. Spur compaction
To: Squeak Virtual Machine Development Discussion
<[hidden email]>



Eliot mentioned using the vmaker sim to load the image and then save
it after removing the empty space. Ah, found it -

> Build a VMMaker image with image/buildspurtrunkvmmaker.sh and you can now use Spur32BitPreen to rewrite the image to be more compact.  We can use this as part of the release process until Clément and I have fixed Spur compaction.  e.g.
>
>       Spur32BitPreen new preenImage: '../oscogvm/image/trunk50'
>       Looking for module  ... loaded...computing accessor depths...done
>       Looking for module  ... loaded...computing accessor depths...done::::.............................................done.
>       old heap size: 41,897,472       initial new heap size: 26,818,472
>       change: -35.99%
>       final new heap size: 26,818,472 change: -35.99%
>
>       Done!
>
tim


Reply | Threaded
Open this post in threaded view
|

Re: Fwd: [Vm-dev] unused-method removal vs. Spur compaction

Ben Coman
Right mailing list. Wrong assumptions. I guessed it would *just*work*.
Thanks for the correction.
cheers -ben

On Mon, Sep 19, 2016 at 3:51 PM, Clément Bera <[hidden email]> wrote:

> I am not sure this is the right mailing list.
>
> Spur32BitPreen could be used as part of the Pharo release process as long as
> the compactor is not working well. This way released images can be more
> compact. If the Pharo folks wants to run Spur32BitPreen from Pharo to
> compact the image, the time spent to make things work may not be worth it as
> the mid-term solution is a new compactor.
>
> On Mon, Sep 19, 2016 at 1:42 AM, Ben Coman <[hidden email]> wrote:
>>
>> Forwarded from squeak & vm lists in case its useful to others...
>>
>> On Sun, Sep 18, 2016 at 5:11 AM, Craig Latta <[hidden email]>
>> wrote:
>> >
>> >
>> > Hi--
>> >
>> >      I've got the mass unused-method removal from Spoon working in Spur
>> > in Squeak 5.1. But now I have a juicy 12-megabyte chunk of free space in
>> > the middle of the 35-megabyte image that won't go away. There's just one
>> > Spur segment (I was expecting several?), with no pinned objects.
>> >
>> >      Is there a known Squeak image with multiple segments? Is there some
>> > way to split a segment? Am I just supposed to play with the grow
>> > overhead and shrink threshold in the VM simulator, until the compaction
>> > I want is triggered?
>> >
>> >      thanks!
>> > Craig Latta
>>
>>
>>
>> ---------- Forwarded message ----------
>> From: tim Rowledge <[hidden email]>
>> Date: Mon, Sep 19, 2016 at 1:17 AM
>> Subject: Re: [Vm-dev] unused-method removal vs. Spur compaction
>> To: Squeak Virtual Machine Development Discussion
>> <[hidden email]>
>>
>>
>>
>> Eliot mentioned using the vmaker sim to load the image and then save
>> it after removing the empty space. Ah, found it -
>>
>> > Build a VMMaker image with image/buildspurtrunkvmmaker.sh and you can
>> > now use Spur32BitPreen to rewrite the image to be more compact.  We can use
>> > this as part of the release process until Clément and I have fixed Spur
>> > compaction.  e.g.
>> >
>> >       Spur32BitPreen new preenImage: '../oscogvm/image/trunk50'
>> >       Looking for module  ... loaded...computing accessor depths...done
>> >       Looking for module  ... loaded...computing accessor
>> > depths...done::::.............................................done.
>> >       old heap size: 41,897,472       initial new heap size: 26,818,472
>> >       change: -35.99%
>> >       final new heap size: 26,818,472 change: -35.99%
>> >
>> >       Done!
>> >
>> tim
>>
>

Reply | Threaded
Open this post in threaded view
|

Re: Fwd: [Vm-dev] unused-method removal vs. Spur compaction

Clément Béra
Well maybe it would just work. One needs to try. 

I am pretty sure we could compress the Pharo image using code run in Squeak out of the box. We may or may not be able to run it from Pharo out of the box.

Another interesting thing could be to use that as part of the bootstrap.

I will try to have a look into the compactor in the incoming week. I teach a lot right now so it's difficult.



On Mon, Sep 19, 2016 at 10:45 AM, Ben Coman <[hidden email]> wrote:
Right mailing list. Wrong assumptions. I guessed it would *just*work*.
Thanks for the correction.
cheers -ben

On Mon, Sep 19, 2016 at 3:51 PM, Clément Bera <[hidden email]> wrote:
> I am not sure this is the right mailing list.
>
> Spur32BitPreen could be used as part of the Pharo release process as long as
> the compactor is not working well. This way released images can be more
> compact. If the Pharo folks wants to run Spur32BitPreen from Pharo to
> compact the image, the time spent to make things work may not be worth it as
> the mid-term solution is a new compactor.
>
> On Mon, Sep 19, 2016 at 1:42 AM, Ben Coman <[hidden email]> wrote:
>>
>> Forwarded from squeak & vm lists in case its useful to others...
>>
>> On Sun, Sep 18, 2016 at 5:11 AM, Craig Latta <[hidden email]>
>> wrote:
>> >
>> >
>> > Hi--
>> >
>> >      I've got the mass unused-method removal from Spoon working in Spur
>> > in Squeak 5.1. But now I have a juicy 12-megabyte chunk of free space in
>> > the middle of the 35-megabyte image that won't go away. There's just one
>> > Spur segment (I was expecting several?), with no pinned objects.
>> >
>> >      Is there a known Squeak image with multiple segments? Is there some
>> > way to split a segment? Am I just supposed to play with the grow
>> > overhead and shrink threshold in the VM simulator, until the compaction
>> > I want is triggered?
>> >
>> >      thanks!
>> > Craig Latta
>>
>>
>>
>> ---------- Forwarded message ----------
>> From: tim Rowledge <[hidden email]>
>> Date: Mon, Sep 19, 2016 at 1:17 AM
>> Subject: Re: [Vm-dev] unused-method removal vs. Spur compaction
>> To: Squeak Virtual Machine Development Discussion
>> <[hidden email]>
>>
>>
>>
>> Eliot mentioned using the vmaker sim to load the image and then save
>> it after removing the empty space. Ah, found it -
>>
>> > Build a VMMaker image with image/buildspurtrunkvmmaker.sh and you can
>> > now use Spur32BitPreen to rewrite the image to be more compact.  We can use
>> > this as part of the release process until Clément and I have fixed Spur
>> > compaction.  e.g.
>> >
>> >       Spur32BitPreen new preenImage: '../oscogvm/image/trunk50'
>> >       Looking for module  ... loaded...computing accessor depths...done
>> >       Looking for module  ... loaded...computing accessor
>> > depths...done::::.............................................done.
>> >       old heap size: 41,897,472       initial new heap size: 26,818,472
>> >       change: -35.99%
>> >       final new heap size: 26,818,472 change: -35.99%
>> >
>> >       Done!
>> >
>> tim
>>
>