From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me
X-Spam-Level: 
X-Spam-Status: No, score=-0.3 required=5.0 tests=BAYES_00,
	REPLYTO_WITHOUT_TO_CC autolearn=no autolearn_force=no version=3.4.4
X-Google-Thread: a07f3367d7,8e11100f675ea2df
X-Google-Attributes: gida07f3367d7,public,usenet
X-Google-NewGroupId: yes
X-Google-Language: ENGLISH,ASCII-7-bit
X-Received: by 10.66.84.98 with SMTP id x2mr7637444pay.10.1357498037053;
        Sun, 06 Jan 2013 10:47:17 -0800 (PST)
From: "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>
Newsgroups: comp.lang.ada
Subject: Re: asynchronous task communication
Date: Tue, 1 Jan 2013 10:59:30 +0100
Organization: cbb software GmbH
Message-ID: <h9l189napmgu$.1h4y5zuvmz6fg.dlg@40tude.net>
References: <1c2dnd5E6PMDR33NnZ2dnUVZ_sednZ2d@earthlink.com>
 <50e18094$0$6583$9b4e6d93@newsspool3.arcor-online.net>
 <7NednS4s2oukfXzNnZ2dnUVZ_oadnZ2d@earthlink.com>
 <aked87Fpfc1U1@mid.individual.net>
 <7cudnYloBfQDw3_NnZ2dnUVZ_rKdnZ2d@earthlink.com>
Reply-To: mailbox@dmitry-kazakov.de
NNTP-Posting-Host: cDN0fd8KlIeJLyErIrSf0A.user.speranza.aioe.org
Mime-Version: 1.0
X-Complaints-To: abuse@aioe.org
User-Agent: 40tude_Dialog/2.0.15.1
X-Notice: Filtered by postfilter v. 0.8.2
Path: 
 6ni88791pbd.1!nntp.google.com!border1.nntp.dca.giganews.com!border4.nntp.dca.giganews.com!border2.nntp.dca.giganews.com!nntp.giganews.com!news.mccarragher.com!news.grnet.gr!de-l.enfer-du-nord.net!feeder1.enfer-du-nord.net!feeder2.enfer-du-nord.net!cs.uu.nl!news.stack.nl!aioe.org!.POSTED!not-for-mail
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
Date: 2013-01-01T10:59:30+01:00
List-Id: <comp.lang.ada>

On Mon, 31 Dec 2012 19:51:36 -0800, Charles Hixson wrote:

> My application logically "needs" thousands or millions of tasks.

An OS can support no more than few hundred task. 

> FWIW, you can think of what I'm doing is a neural network simulation.

Well, it is a kind of structure (data-driven machine) which is very
difficult for concurrent computing based on traditional architectures. Even
if you could have a task per node/cell, the performance could be actually
worse due to task switching, which would eat most of the resources.
 
> Since I can't model cells as tasks, I'm planning on modeling them as 
> protected variables.

I would not do that. There is a possibility that protected actions would
use system-wide locks. And it is a very bad idea to do perform anything
time consuming (e.g. calculations) from a protected action. Logically a
protected action is considered "instant."

For a NN, if you really wanted to experiment with concurrency, I would use
a pool of tasks (there already were good advices how many of them make
sense to have on a multi-core). The tasks would service groups of nodes of
the same layer. For data you would need no locking at all, since the layer
N+1 would not start before the layer N is completed.

For a completely data-driven architecture I would again recommend a pool of
tasks looking into a blackboard for requests [lock-free]. As I said there
is a fair possibility that you would actually lose more than gain, but you
could try. [lock-free FIFO is for 1-1, blackboard for 1-n producer-consumer
scenarios].

Happy New Year to all,

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de