Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!mit-eddie!gary
From: gary@eddie.MIT.EDU (Gary Samad)
Newsgroups: comp.sys.amiga
Subject: Re: Does RAM: ever retry?
Message-ID: <6756@eddie.MIT.EDU>
Date: Thu, 3-Sep-87 16:45:02 EDT
Article-I.D.: eddie.6756
Posted: Thu Sep  3 16:45:02 1987
Date-Received: Sat, 5-Sep-87 11:33:33 EDT
References: <510@sugar.UUCP> <6619@eddie.MIT.EDU> <535@sugar.UUCP> <7069@steinmetz.steinmetz.UUCP> <566@sugar.UUCP>
Reply-To: gary@eddie.MIT.EDU (Gary Samad)
Organization: MIT, EE/CS Computer Facilities, Cambridge, MA
Lines: 37

In article <566@sugar.UUCP> peter@sugar.UUCP (Peter da Silva) writes:
}I still think it's gross...
}
}> The 1.2 Ramdisk has a 'ripcord' of 30K of ram.  If it finds it can't get
}> enough memory, it releases it's 'ripcord' and puts up the disk full
}> requestor.   It won't agree to be un-full until it can get a ripcord
}> back.  This was to solve problems of copying too much to the ram disk
}> and having the system crash.  Without the 30K, it's hard to let the user
}> know the ramdisk is full and let the user correct the problem.
}
}But does that 30K have to be contiguous? And what's wrong with just saving
}enough for the requestor? Surely a system requestor can fit in less than 30K.

'Fraid not.  During development of Microfiche Filer I spent several full
days writing routines to handle low memory conditions.  The major task
was to throw up a requestor alerting the user whenever this happened.  So,
I computed that I should need about }i10K for my requestor (size of window
* 2 bitplanes * 2 again toover the smart bitmaps that are allocated
when you open a window on top of a SMART_REFRESH window).  "No problem,
just set up a small ripcord," thought I.  "10K should do it."  "Whoops,
well how about 20K?" "30K?" "Ah, there's the requestor." "But not every time.
How about 50K?" "Nope, 60K?"

It took 60K to be able to RELIABLY open my requestor!!!!!  Even though the
computed space needs should have been around 10K.

My solution?  Give the user a "memory paranoid" option which opens that
window and just keeps it around, but at the bottom of the window pile.
Yes, this could be a tacky solution but it works fine for my particular
product because I always have a bigger window around, behind which I keep
this "paranoid" window.  It works beautifully.

But why does it actually require up to 60K to open a requestor that only
takes 10K after it is open?  (I have verified that it only takes about
10K once it is opened.)

	Gary