[Python-ideas] An alternate approach to async IO

[Trent Nelson]

On Tue, Nov 27, 2012 at 12:54:50PM -0800, Greg Ewing wrote:
> Trent Nelson wrote:
> >     So what about the "no processing that can be usefully done
> >     by a C level thread" bit?  I'm trying to discern whether or not you're
> >     highlighting a fundamental flaw in the theory/idea ;-)
> 
> You seem to be assuming that the little bit of processing
> needed to get the data from kernel to user space is going
> to be significant compared to whatever the Python code is
> going to do with the data.  That seems unlikely.

    Great response.  Again, highlights the need to have some
    standard way for benchmarking this sort of stuff.  Would
    be a good use of Snakebite; everything is on a gigabit
    switch on the same subnet, multiple boxes could simulate
    client load against each server being benchmarked.

    (I don't think my idea would really start to show gains
     until you're dealing with tens of thousands of connections
     and 1Gb/10Gb Ethernet traffic, to be honest.  So, as I
     mentioned in a previous e-mail, it depends on what our
     goals are for the next-gen AIO framework (performance
     at whatever the cost vs not).)


        Trent.