>> My question is, does the current VSTa system allow high-level servers to take
>> lots of CPU running the requests of low level clients? If it does, then it
>> certainly does not "fairly" share the CPU. QNX has a system for adjusting
>> the priority of a server to match the priority of the client whose request
>> it's serving. Is something like this worth looking into?
>
> This sounds like something that would need a large scheduler overhaul. I
> think going this way we'd also hit issues like whether or not we servers
> should reorder their queued clients (I think I'll leave that to someone else
> just now :-)).
Yes, I think priority queueing would be something which would need to be
added to provide an appropriate mechanism to make this type of system work.
However, even without priority queueing. A multithreaded server which
adjusted each thread's priority according to the priority of the client
which requested the work (one work request per thread) would provide an
EXTREMELY "fair" distribution of CPU resources according to priority.
However, it does sound like something which can be left until later.
Ideas like this is why I original rementioned the idea of integrating
external scheduler loading (even if it was into kernel space) because
it would provide an easy way to work with this type of expansion. However,
either of these are things that I agree would best be left until later.
Received on Wed Dec 14 03:35:25 1994
This archive was generated by hypermail 2.1.8 : Thu Sep 22 2005 - 15:12:11 PDT