> -----Original Message-----
> From: Ann W. Harrison [mailto:aharrison@...]
> Sent: Lunes 26 de Febrero de 2001 16:40
>
> We need to work on threading and on effective use of large cache
> sizes if we are to handle thousands of concurrent attachments.

I've seen a daunting structure that's used to keep track on how much a
request has been run. If the time tracked is more than a threshold, the
current execution gives up in favor of another request being executed. In
this context, "give up" is to pass the control to a peer thread, not to
abort the thread that reached the threshold.
This manual handling of threads works at the same time than the thread
scheduling done by the operating system. Is this observation correct? Could
we get rid of such mechanism in the future?
At least in NT, we have two options:
a) Get rid of manual thread scheduling and let the OS to manage or mismanage
the threads at will.
b) Leave the manual scheduling in place and put conditional directives to
use fibers instead of threads. Fibers are lightweight threads were the
process is in charge of managing their quantas, not the OS. MsSql has this
option.
Of course, the most general route would be a) that would avoid at the same
time to put more platform dependant code.

C.