I don't remember if the OP specified in what respect the suggested 
implementation should be efficient.  It could be efficient in terms of elapsed 
run time, CPU time used, average virtual storage working set size, average real 
storage working set size, or programmer development time (presumably the OP).  
The suggestion I made, for the 512MB bit string, maximizes the efficiency of 
the programmer developing the scheme.  Other suggestions seen so far emphasize 
some of the other four goals listed above.  I personally am far more concerned 
these days with my own development time than I am with execution time 
efficiency.  45 years ago, the reverse was true.  The only exception now is 
when my piece of code must execute one or more million times per second with 
interrupts disabled, holding a lock, or in some other exotic state.

The bit string scheme will require 512 MB of virtual storage, but far less than 
that in real storage backing the virtual storage working set, as the OP has 
himself indicated that the total number of input items, each of which could 
cause a separate page fault and short term page fixing, is far smaller than 
128K, which is the number of 4K pages in 512 MB.  I was flattered when William 
Blair experimented with my idea.  He reported that one of his worst case 
scenarios took 7 seconds of elapsed time.  Only the OP knows if 7 seconds is 
acceptable.

Whenever someone asks for efficiency or performance to be optimized, my first 
question is "with respect to what variable"?

Bill Fairchild
Rocket Software

-----Original Message-----
From: IBM Mainframe Assembler List [mailto:[email protected]] On 
Behalf Of robin
Sent: Wednesday, August 25, 2010 6:25 AM
To: [email protected]
Subject: Re: Efficient Memory List

From: "Mike Shaw" <[email protected]>
Sent: Wednesday, 25 August 2010 12:25 PM

> On Tue, Aug 24, 2010 at 1:40 PM, Patrick Roehl 
> <[email protected]>wrote:
>
>> <snip>
>> Because of these factors, and the ideas brought out in this discussion,
>> the bit string idea seems like an excellent fit.
>
> Patrick:
>
> If you implement the bit string idea, please let us know how much of an
> improvement (execution time, CPU time) it provides over your current exit.

Plus also the effect that the scheme has on the performance of other running 
programs,
when you request and use some 500Mb of storage.

Reply via email to