Hi
I forward you the email Andreas wanted to send to the list.
Here it is.
Stef
From: Andreas Raab <[email protected]>
Date: December 19, 2008 6:41:21 AM CEST
To: Igor Stasenko <[email protected]>
Cc: Oscar Nierstrasz <[email protected]>, stephane ducasse <[email protected]
>, "Klaus D. Witzel" <[email protected]>, Alexandre Bergel <[email protected]
>, [email protected], Eliot Miranda <[email protected]
>
Subject: Re: [Pharo-project] Multi-core Pharo?
Hi Igor -
I think I'm missing some context here. I don't know what sore spot
Oscar is referring to and which political problems. From my
perspective the situation is that we have released Hydra under MIT and
that consequently no "political problems" exist (at least on this end).
From the technical perspective you are of course right. The issues
need to be addressed from various angles simultaneously. Hydra at this
point is a proof-of-concept, no less, no more. To make this useful we
need to address a variety of problems, including how to actually write
programs in such a setup! ;-) This gets forgotten too easily - people
assume that things get magically faster but really, Eliot is working
on that part. Hydra requires some deep thinking about to structure
applications properly so that they actually perform better than before.
Of course, having smaller images will help tremendously because it
allows more experimentation with smaller footprints. This is an issue
that we continue to look at both from the shrinking as well as the
construction approach.
Cheers,
- Andreas
Igor Stasenko wrote:
2008/12/18 Oscar Nierstrasz <[email protected]>:
I didn't understand a thing :-(
I seem to have touched a sore spot, but I do not know what the
background
is.
Can someone summarize? Technically and politically, what is the
current
status of Hydra, and what are our options for using it as a basis for
multi-core Pharo?
First, I think this discussion missing two important persons: Eliot
and Andreas.
Andreas is one, who behind the idea of Hydra VM. Eliot is making
serious rewritings of squeak VM (namely Cog).
There are plans to merge Hydra & Cog and continue with a new VM.
This is beneficial in many aspects: i really think that best for all
of us is to join efforts to produce better VM, instead in having
multiple branches Squeak/Cog/Hydra VMs.
Currently, works on Hydra is suspended because the more changes i do,
the harder it would be to merge with Cog.
Merging mess of low-structured low-level code of VMMaker is tough
task, so i don't want to make it harder. :)
Moreover, Hydra initially was proposed as experiment initiated by
Qwaq, implemented by me. It turned out to be successfull one. :)
But to leverage it, we need a much more than just VM - we need people
who think/know how to adopt things to get benefits from
multithreading.
Squeak images simply not ready for this. We should have a better ways
to scale: instead of loading bloated images (each for own interpter),
we should think how to lessen the bloat and use single central image,
which then managing smaller/nimble ones.
That's why i keep making noise about modularity: we need modules which
can be easily installed and run more or less standalone without
cross-package dependencies.
There is nothing wonderful in Hydra alone: we should attack the
problem from both ends - language side and VM.
- on
On Dec 18, 2008, at 10:08, stephane ducasse wrote:
Ok this is clearer now.
Stef
On Dec 18, 2008, at 9:40 AM, Klaus D. Witzel wrote:
On Thu, 18 Dec 2008 09:13:11 +0100, Stef wrote:
Klaus
if you want to have a clear answer from esug:
You're mixing up things here. If Pharo people want to run Hydra
on their
Mac, they could approach ESUG for funding.
- follow the process (cf the web page)
- write a clear request
Since you (not me) would be the benefitor, go ahead. I was only
the man
who suggested the combination. Don't blame the messenger, don't
put your
work on him. Is this clear now?
So far it was a little mail about moebius and not Hydra.
Yes, clear. But it is the same developer, namely Igor, who would
do that
for Pharo Mac users.
Cheers,
Klaus
Stef
PS: ESUG has a board and we discuss and vote before taking
decision.
Really? ;) to me you seem to be more concerned about your own
political
attitude than about getting Hydra to the Mac/Pharo community.
Please take notice that I know precisely how the democratic
process in
the ESUG board is supposed to work, no lecture necessary from
your side. You
are wasting time (not the first time with me).
On Dec 17, 2008, at 7:58 PM, Klaus D. Witzel wrote:
Hi Alex,
I'm sorry I'm just a frequent reader of the Pharo mailing list,
so
here's my response by email. Feel free to forward it.
On Wed, 17 Dec 2008 15:11:52 +0100, you wrote:
Hi Igor and Klaus,
Having a description about Hydra will really help promoting your
project.
Maybe we could continue what I started to write...
The Hydra VM is usable as is and in production, serving heavy
used
Croquet site, IIRC. Hydra has everything you need for the Multi-
core VM of
Squeak:
- launch and control multiple .images in native threads
- control these threads from the Smalltalk language level
- communicate beween .images through channels
- any number of related/unrelated .images can be run
What is not done yet (Igor would correct me) is making Hydra
available
on Mac and Linux(en).
In light of recent discussion with Stef about ESUG founding, I
would
suggest you perhaps consider to make Hydra available for you
Mac users.
Cheers,
Klaus
P.S. Alex the "rest" of the discussion you recalled above (the
paper
you suggested) is about a very particular use of Hydra
technology, formost
for and by enthusiastic VM engineers to be "the first" who get
it going.
Hydra can do and does well without all that (without the
cloning of object
subgraphs "baby Hydras" from the running .image): just save
an .image file
and then start it up in Hydra.
Alexandre
On 17 Dec 2008, at 09:52, Oscar Nierstrasz wrote:
Hi folks,
Has anyone looked at the Qwac Hydra stuff?
http://www.qwaq.com/HydraVM/
http://lists.squeakfoundation.org/pipermail/squeak-dev/2008-February/125719.html
There was quite a lot of discussion on squeak-dev ast
February, but
not much since then.
I would really love to see a multi-core Pharo. If someone
has an
idea
how this could be done, I would be interested to learn more.
I think
getting Pharo to exploit multi-cores could be a big selling
point in
the future ...
- on
_______________________________________________
Pharo-project mailing list
[email protected]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
_______________________________________________
Pharo-project mailing list
[email protected]
http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
