[geeks] PS/3's, Linux, etc, was: Second Life is not a game?

Wed Aug 1 15:03:42 CDT 2007

On Wed, Aug 01, 2007 at 07:31:22PM +0100, Mike Meredith wrote:
> Now imagine that some company has a big cluster consisting of dozens
> (or more) of Cell blades. Do you let the developers have access to that
> cluster when access to a PS3 would do almost as well ? You may need to
> let the developers use the cluster for some things, but quite a bit can
> be accomplished with a PS3.

I would not do it that way. (differening styles), I would have a smaller
cluster of Cell blades for development and testing. 

My style of development would be to develop the algorythms on a large
PC one per programmer. Since most of the time is spent in editors and
debuggers running single threaded, it's not going to matter, except for
the endianness what they run on. 

Once they need to go to a real Cell system, a cluster for development
makes a lot more sense to me. It's cost can be shared among the programers,
new software can be tested on it without affecting the production cluster
and if need be it can be reduced by a blade if a production one fails.

If things really go bad, the development cluster  could be made the production
one in extreme cases. 

I am prejudiced against PS/3's because they are really made to be toys.
Expensive, high end toys, but toys none the less. I would be much happier
using more reliable COTS (commerical off the shelf) hardware. 

> I'm not talking about a developer being stuck in front of a wide screen
> TV with a bluetooth keyboard hacking code on the PS3. Just attach a PS3
> using a cross-over cable to the back of the developer's workstation.

That sounds like a reasonable plan, except I would have it on the 
corporate network so that it can be backed up, etc. I do a lot of work on
remote computers using terminal windows, I happen to be sitting at an XP
computer running putty, connected to a Linux system.

I'm never sure about saving money by skimping on hardware. While the
economy sucks here, the best LAMP programmers and kernel experts are
making 35k  NIS (around $8k) a month, plus benefits and my guess is
now that demand is heating up, they will be up to 50k ($12k) in a year.

At that price, I had better be able to afford to buy them a really good 
computer or I am wasting my money. 

> As for a lone geek using a PS3 as a workstation and trying to make
> code like _oggenc_ faster using the 'funky cores' *, well *I* wouldn't
> do it but good luck to him. A PS3 (or a hypothetical PS4) isn't going
> to replace my Ultra40 but peering into the future I don't see it as
> impossible that my next workstation could have something like the Cell
> in it ... perhaps AMD's vague plans to include non-x86_64 cores (as
> well as x86_64 cores) in their multi-core processors will include a
> number of Cell-like 'funky cores' to speed up certain operations like
> MP3-encoding, etc.

I don't know how that would apply in the real world. Most people don't
do mp3 encoding enough to care, but real time video compression could
be a hot item in the thousands of little shops that do video editing.

Believe it or not, there really are algorythms that compress full motion
video on small screens down to about 64k bits per second and can be
decompressed with a cell phone grade processor. Cringly wrote about it
a year or so ago, look him up, don't ask me for citations, I was handed
the article on paper and it's gone.

That would do well on a Cell type processor, especialy for live feeds 
to handheld devices. 

I smell a start-up :-)

> *: This is a *real* bad comparison, but my PS3 seems to encode an album
> to MP3 *much* faster than Ultra40 will. I should probably get some
> timings!

May not mean a thing. Once you get beyond 256k, I/O bandwith and other
issues apply. It may really be all down to the audio ripping speed of
your CD drive. :-)

> I'd assume that to make money with GPLed software one would sell
> support rather than software. For some markets this makes sense ...
> where the initial license for the software is dwarfed by the ongoing
> support costs (something like student records systems for
> Universities ... where each University makes changes to the
> 'off-the-shelf application' and pays support from the vendor). 

Yes, that's the real money maker. Actually it's a combination of what
I call agreegation, putting things together that need to be a unit
for example in the games world, a game and a emultator etc, distribution,
(delivering it to the customer) and support.

> I'm not really much of a developer and excluding some minor patches for
> ISC's dhcpd, haven't contributed much code anywhere. But yes I'd very
> very wary of contributing code to a project like that.

I was not involved with the project at all, but I see a market for one
project where that happend for a closed source drop in replacement. I
figure that most of the outside people who did the really good work and
left the project when that happened would love to come work for a
company that rewarded them fairly for their work and also supported a
seperate GPL'ed project.

The aims of the two projects would be different. The closed source one
would be designed for reliability and performance while the GPL'ed one
would be designed for ease of extension, and new functions would be 
encouraged. 

Since GPL'ed code can't be patented, the ideas used in it could be used
with seperately written code in the proprietary version. That way everyone
wins. :-)

Geoff.

-- 
Geoffrey S. Mendelson, Jerusalem, Israel gsm at mendelson.com  N3OWJ/4X1GM
IL Voice: (07)-7424-1667 U.S. Voice: 1-215-821-1838 
Visit my 'blog at http://geoffstechno.livejournal.com/