[unios] Re: Generic Design

Anders Petersson anders.petersson@mbox320.swipnet.se
Thu, 10 Dec 1998 21:45:31 +0100


From: Anders Petersson <anders.petersson@mbox320.swipnet.se>

At 20:31 1998-12-10 , you wrote:
>From: Pieter Dumon <Pieter.Dumon@rug.ac.be>
>
>Some more comments on the real content of your document. 
>I won't comment on the principles, allthough I'm not convinced.

Not convinced? That was a mild statement.

>> Hardware
>> --------
>> ...
>> There are exceptions to the rule that hardware is objects. One is storage,
>> which isn't used in the same way, and should be protected from being used
>> up in some other way. 
> 
>Do you recon that storage is a big part of the devices in a computer?

The actual devices are objects. The space itself is not. That's what I
meant. I can't think of any good way to make the storage room into objects.
This should be better protected by limiting each user to a certain size of
storage or something.

>>CPUs aren't objects either. Objects shouldn't deal
>> with the CPU(s) at all. Their only worry is to get enough processing time,
>> which is handled as concrete things that can be possessed. Realtime
>> processes get a real value, whilst usual processes get a percentage of the
>> availible processing power.
>
>If you would try to implement such a schedulign algorithm, it would kill
>you system. Just work with multiple priority queues. Real-time processes
>and drivers get high, static priority, normal processes get a dynamic, low
>priotity. Dynamic means it is changed by the way the process behaves: 
>if the process uses up the time quantum that is assigned to it, its
>priority decreases, if it yields control to the system before it's time is
>used up, it's priority increases. Lower priority queues get a longer
>time quantum.  In this way, interactive processes, i.e. processes that do
>I/O or regularily wait for some event, get a higher priority but can
>execute shorter. cpu-intensive processes get a lower priority but can hold
>the cpu longer. This is roughly algorithm used in all OSs. Why? Just
>because it closely resembles the mathematically best scheduling algorithm.

This was actually more of a loose thought. You're probably right that this
would be too slow. But I still think processing time has to be delegated in
a controlled way. One user can't take up all computer power if others don't
want him to.

>> Processing power is delegated down the system
>> tree. The root owns all time. Each user maybe owns an equal percentage of
>> the time. The users may then give their respective time shares to processes
>> as they like. The same method could be used for all kinds of storage.
>> Otherwise I can't think of any hardware device that shouldn't be an object.
>
>This would really hurt the system! Users who only run one command-line 
>UI don't need much power !!!!

Some users don't need much power, others need every drop of power
availible. Don't forget about the 'OS for everyone'. My point here was the
same as the previous one - that one user can't hog all processing time.
It's very possible that this could be made more efficient... suggestions
are welcome.

binEng


------------------------------------------------------------------------
Help support ONElist, while generating interest in your product or
service. ONElist has a variety of advertising packages. Visit
http://www.onelist.com/advert.html for more information.
------------------------------------------------------------------------
UniOS Group
http://members.xoom.com/unios