youhou oscit just got slower !

[Gaspard Bucher]

With great efforts, lots of reading and clever coding, I managed to
slow down oscit library from 9s execution to 12s during the execution
of the 392 tests !

;-)

It's really slower, but for a good cause: thread-safety. I did not
want a design with a global lock: adding threads to objects or
creating new processing flows in rubyk should be *natural*, and there
should be no way to crash an oscit application because of this.

The slow down might look impressive but it is only happening in
regions of the library that are tested a lot but not used often during
signal processing (we are testing oscit, not rubyk).

In order to reduce the overhead and the bottlenecks related to
concurrency, I used atomic counters for reference counting,
readers-writer locks and tiny scoped locks.

Let's look at this overhead has an investment for the future: with
thread-safety and a good concurrent design, we will be ready for the
massive multicore processors.

Goodnight !

Gaspard

[Quentin Berthet]

Good job !
It's nice to see software writen from the ground up with fiability and
concurrency in mind. But as you said, physics's limits will not be
pushed far more than 4 or 5 ghz, so go go multicore :-)

On a side note, you are probably aware that apple will announce his
"tablet?" mac tomorrow. I see in this, in conjonction with rubyk/oscit/
mimas, a good way to build a much cheaper lemur ... (the ipad should
be 10.1" multitouch, same as lemur).
Also if they use the same business model than for iphone, this could
be a good way to make money.

wait impatiently & see carefully

Bye
Quentin

Le 26 janv. 10 à 02:07, Gaspard Bucher a écrit :

> --
> _______________________________________________
> rubyk mailing list
> ru...@rubyk.org
> http://rubyk.org/en/community

[Gaspard Buma]

I thought mimas with multi-touch interfaces in mind:

* either you do everything with the keyboard (for the vim adicts)
* or you should be able to work, just with fingers (no double-click,
no "hover", no keyboard entry, etc)

Thanks for the feedback !

Gaspard

On Tue, Jan 26, 2010 at 9:56 PM, Quentin Berthet <tum...@gmail.com> wrote:
> Good job !
> It's nice to see software writen from the ground up with fiability and
> concurrency in mind. But as you said, physics's limits will not be pushed
> far more than 4 or 5 ghz, so go go multicore :-)
>
> On a side note, you are probably aware that apple will announce his
> "tablet?" mac tomorrow. I see in this, in conjonction with

> rubyk/oscit/mimas, a good way to build a much cheaper lemur ... (the ipad