[Stackless] Performance and slprofile Re: Comments on a Twisted/Stackless Integration Example
cstejerean at gmail.com
Tue Jul 3 23:18:14 CEST 2007
Tasklets are probably not useful for concurrent number crunching, if
each tasklet is CPU bound then you are probably better off with
multiple processes on multiple CPUs. If tasklet are doing async IO
however then you probably are not CPU bound and can continue adding
tasklets I would argue until you saturate your CPU.
Cosmin Stejerean (m)
On Jul 3, 2007, at 4:08 PM, Andrew Francis <andrewfr_ice at yahoo.com>
> Hi Arnar:
> --- Arnar Birgisson <arnarbi at gmail.com> wrote:
> AF> 1) An extra tasklet is created for each
> AF> connection.
> AB> I think that's pretty much the standard way of
> AB>doing things. The beauty of tasklets is that
> AB>they're cheap so you shouldn't worry about
> AB> having too many (tens of thousands is fine).
> I understand tasklets are cheap but still there are
> costs. For example given the round-robin scheduling(I
> figure this is an O(n) algorithm), the more tasklets
> one adds, it takes that much longer to service a
> tasklet. This may be fine if one has few runnable
> tasklets and connections. However what happens if one
> has hundreds of connections? When does the scheduling
> and processing costs add up to a point where a tasklet
> cannot return a response in a reasonable amount of
> When I have time, I would like to figure out how to
> use slprofile so I can understand what is happening
> under the hood. If anyone can give pointers, that
> would be great.
> Looking for a deal? Find great prices on flights and hotels with
> Yahoo! FareChase.
> Stackless mailing list
> Stackless at stackless.com
Stackless mailing list
Stackless at stackless.com
More information about the Stackless