Subject: | |
From: | |
Reply To: | |
Date: | Fri, 19 Oct 2001 11:06:11 -0700 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Re:
> The performance enhancements achieved by the use of these files is not
> so much the result of them being memory-resident (or not), but rather
> because they can be searched by a binary search algorithm using
> memory-mapped I/O techniques. In benchmark testing I did, doing lookups
...
Wow...that recalls memories from the past! We did something like that
for a school back on the Classic HP 3000, around 1985 or so...
and got about a factor of 60 speed improvement. (I recall that
we averaged 1/2 millisecond with our lookup vs. 30 milliseconds for an
IMAGE DBFIND/DBGET.) In our case, we used a data segment to hold every nth
record, and then flat files to hold every record. A binary search of the
data segment told us what file to then binary search. (Today, of course,
we wouldn't use data segments :)
<plug>
If you want to determine if a file is in memory, the KLONDIKE tool from
Lund Performance Solutions will show you:
:klondike count foo.pub.sys
foo.pub.sys @ $699.0: 21 pages; InMem: 0 (0.0% of file, 0.00% of memory)
and, it can fetch it into memory for you:
:klondike fetch foo.pub.sys
21 pages; InMem: 21 (100.0% of file, 0.03% of memory); 21 Referenced
</plug>
Stan Sieler [log in to unmask]
www.allegro.com/sieler/wanted/index.html www.allegro.com/sieler
* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *
|
|
|