On Thursday 21 March 2002 11.11, Lincoln Dale wrote:
> as for [1], SIGIO sucks. if you really want to know the gory details why,
> see http://marc.theaimsgroup.com/?l=linux-kernel&m=100888541205739&w=2
> http://marc.theaimsgroup.com/?l=linux-kernel&m=100888593806860&w=2
> http://marc.theaimsgroup.com/?l=linux-aio&m=100888973615146&w=2
I know enought of it to say that the Linux approach to SIGIO (use of RT
signals) do not suck, only some "minor" details of the acutal implementation
sucks.
SIGIO in general do indeed suck badly by the design of the event delivery
mechanism...
> as for [2], its only solving part of what needs to be done. once you have
> scalable event-notification and you've fixed up disk i/o bottlenecks,
> you'll then discover that your next bottleneck is that you're running out
> of memory bandwidth due to the large number of memory-copies that go on for
> every read()/write(). at that point, you'll see that you need some other
> mechanism.
Exacly.
> if you're really interested in following this, take a look at linux-aio ...
AIO do not really solve all of these problems either, but in principle the
approach is mostly correct, and the problems can probably be fixed within the
framework given some time and cleverness (not that the result will really
resemble AIO by then...)
Regards
Henrik
Received on Thu Mar 21 2002 - 07:09:57 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:14:52 MST