rsingh1%dahlia.waterloo.edu@cunyvm.cuny.edu (12/16/89)
Interesting BENCHMARK I did at the end of file that rates
the speed of the amiga's hashing file system. Probably not too accurate
for what it's measuring, but maybe it is. Check it out.
REply:
You said that MS-DOS beats the amiga out on it's handling of
directories.
For SMALL directories, this is ABSOLUTELY true.
-> B | | T <-
|_|
There are LOTS of bad points!
For example: If the part of the disk that holds the directory gets
wiped out, it's game over. No way to recover. Sure, there are UTILITIES
that create backup's and things, but the problem still remains.
ALSO! -> Have you ever seen MSDOS in action with a LARGE directory?
(a few hundred files). Slow ISN'T THE WORD. The speed is remarkably
pathetic. The 'simple approach' of searching sequentialy is seriously
dumb. And outrageously slow. On most large IBM BBS's that rely on
the files in the file section as data, things slow down to a crawl.
Look what happens when Opus gets more than 500 or so messages in an
area. And look how long it takes QuickBBS to do a new files with a large
file section. The problem was so bad that one sysop split his GIF section
into ALPHABETIC sections. Even though we still do a new files on the same
# of files, the speed has increased by a few orders of magnitude.
Those are the main points. I have worked on databases with thousands of files
, and it really is a pain to split databases, because the system's file
system becomes a bottleneck.
The hashing system is a good deal more complex, but boy is it fast.
Especialy for pattern matches and stuff. Here are some benchmarks
on a semi-full directory running off of floppy:
Files: 71 with real file names, and 100 with random file names.
The benchmark program was written in Arexx, and looked like this:
/* Timer.Rexx -- Times how long a command takes to execute */
a=1
address command
call time 'R'
'dir >nil: df1:shit'
call a
call time 'R'
'dir >nil: df1:shit/(S*|*s|*d|*me*|[a-z]??[n])'
call a
call time 'R'
'dir >nil: df1:shit/(S*|*s|*d|*me*|[a-z]??[n])'
call a
call time 'R'
'dir >nil: df1:shit/?????'
call a
call time 'R'
'dir >nil: df1:shit/~(s*)'
call a
exit
a:
Say 'Test #'a 'with an elapsed:' time('E')
a=a+1
return
The purpose of this benchmark isn't to show how slow the seek access
times on the floppy are, so I will run it with a cashe (most IBM
systems run one too it seems. The cashe being used is called
FACCII, and looks to the computer, about the same as the drive.
The line: 'dir >nil: df1:shit/(S*|*s|*d|*me*|[a-z]??[n])'
will list everything that begins, and ends with s, what ends with d,
what contains the letters 'me', and then it lists
every 4 letter file that begins with a letter and ends in n.
The line: 'dir >nil: df1:shit/?????'
simply lists every 5 letter file.
The line: 'dir >nil: df1:shit/~(s*)'
simply lists every file that doesn't start with s.
The output is all disgarded in the nil: device. Also, understand that
all of the output would be sorted.
Benchmark Begin: (Direct snip from console window)
6> rx timer
Test #1 with an elapsed: 2.72
Test #2 with an elapsed: 2.26
Test #3 with an elapsed: 2.26
Test #4 with an elapsed: 2.02
Test #5 with an elapsed: 2.56
All results are in seconds. With 171 files, and a terribly fragmented
directory (although that shouldn't matter in these benchmarks), the
speed isn't bad. And the pattern matching is quite complex.
I hope I've proved that the amiga's file system is a good deal better
than the ms dos file system. The amiga's file system excells in finding
a file with a particular name quickly, and it can perform quick pattern
matching. Also note, that this was with the 'sfs' (slow file system),
and that about .1 of those seconds are used to call up the directory
command.
-paul sop