Re: [hybi] frame length encoding

["Shelby Moore" <[e-mail]>]

> On Sun, Aug 22, 2010 at 12:00 AM, Shelby Moore <[e-mail]>
> wrote:
>
>> Why can't you utilize more than one core?
>
> There is no effective way to turn "process the frames on this
> connection" into a concurrent process.

Actually just remembered I had this public discussion last year in design
of Copute, and the conclusion is you can't backup the iteration and still
be pure (or something like that, I will go look at my notes again).

The channel is serial and forks (side-effects), so you can't parallize it
unless you are willing to do predictive branches and throw away CPU time
for incorrect guesses.

Okay I think understand.

> Yes, there are many very slow networks around, just like there are
> many devices with limited memory.  However even those slow networks
> get consistently faster.  Just because GPRS is not 10GigE does not
> mean mobile users won't get LTE upgrades.

But it seems subjectively that the rate speed is introduced at the
bleeding edge, is running away from the rate that speed is improving at
the dull edge.  But perhaps I am wrong about that.

I am highly frustrated that afaik, I can not buy a faster pipe here for
any price.  My solution was to buy multiple crappy pipes and load balance
them :D

I think it is economics over here, so I tend to think you will be proven
correct.  Once the dollar and the western nations collapse by 2011 or
2012, we will see massive boost in relative wealth in the developing
world, and then a massive increase in the bandwidth in the consumer
markets here.  Time to get positioned now for the big shift (are you
listening Google?).


> Shelby, if you are interested in discussing parallel architectures and
> high performance networking, we can talk off-list.  This is not
> relevant to HyBi any longer.

Is that an interest of yours?  Okay I may hit you up later.  Thanks.

>
> Regards
> -
> Pieter Hintjens
> iMatix - www.imatix.com
>
>