Black Box 4/BlackBox.cdr/finance/magev40c.arj/DOCS.ZIP/SPEED.DOC

Text File | 1991-07-06 | 12KB | 200 lines

The DATAMAGE data management system:  Comments on execution speed.

Speed  of  execution  is  a topic of interest  in  any  program  that  performs
significant tasks.  Databases can be said to live or die by their ability to do
MASSIVE  processing  in  literally no time flat.  Many  of  the  programs  that
compete  with  DATAMAGE  waste incredible amounts of  disk  space  and  include
methods that render their files and indexes unimpregnable to any other  program
in order to accomplish, or to SEEM to accomplish, this impossible goal.

There  are ways to speed things up, and it is the purpose of this  document  to
explain  them.   These methods fall into three basic categories:   Get  a  fast
computer,  use the computer's memory instead of the disk drive, and  keep  your
files all in one place on the disk drive.

HARDWARE:

Certainly,  the  raw  power  of  the target  computer  is  a  very  significant
ingredient  in  the  equation.   If  you are using an  old  PC  or  PCXT  whose
microprocessor is the original 8088 and runs at 4.77 Mhz you will be quite dis-
satisfied  with the performance of this, or any other program.  (A Mhz, by  the
way,  is  one million clock cycles per second.)  If you can not  afford  better
hardware one thing that will enhance your current computer is to pull the  8088
chip  out  and  replace  it with a NEC V-20.   This  will  almost  double  your
processing speed but your computer will still be, in modern terms, VERY SLOW.

The  original PCAT, running at 6 Mhz, and the later "enhanced" version of  same
running  at 8 Mhz were fast machines in their day.  Their time has  passed.   A
modern  AT-class  machine should run at a minimum of 12 Mhz,  and  hardware  is
available  that  runs  all  the way up to 44 Mhz!   There  is  little,  if  any
difference  between  an 80286, 80386 or even the latest  80486  microprocessors
when  shuffling  datafiles.  The CLOCK rules the computer,  and  dictates  it's
speed.  No other enhancement can approach the clock speed for doing it fast.

SUBSTITUTING MEMORY FOR THE DISK DRIVE:

Memory  above MS-DOS's 1,024,000 byte (1 MEGABYTE) limit is available  ONLY  on
AT-class machines, or with special E.E.M.S cards on XT-class machines.   MS-DOS
can not use this memory, but your computer can make use of it via DRIVERS which
switch  the microprocessor into it's native mode, make use of the memory,  then
switch  it  back into 8088 emulation mode and continue to execute  the  program
that  runs  under  MS-DOS.  That's right - when you  are  running  MS-DOS  your
microprocessor,  be  it  a '286, '386 or '486, is NOT  running  it's  potential
instruction  set, but is emulating an 8088.  The newer microprocessors CAN  NOT
run MS-DOS, nor can MS-DOS access or manage memory over 1 megabyte.

There  are two types of drivers that can add significant speed to DATAMAGE,  or
any other program that makes heavy usage of the disk drive:  RAM-DISKS and DISK
CACHETING.  These drivers allow the computer to use memory above the 1 megabyte
limit and substitute it for the slower disk drives.

With the RAM DISKS a "fake" disk drive is added to your computer.  You can  use
it  just as you would any real drive, you can do everything except  remove  and
replace  the disk, which is also not possible with a hard disk.  DATAMAGE  will
prompt you for the disk drive to use for various tasks.  If you have sufficient
memory you can set up a ram-disk and greatly profit in terms of speed.

With  the DISK CACHETING drivers the data in the files currently open  on  your
computer is read from the disk where it resides and moved into memory above the
1 megabyte limit.  When you read or write data it is moved from/to this  memory
which takes quite a bit less time than doing the same operation from the disk.

BENCH MARKS:

Before beginning it may be helpful to explain that the standard benchmark of PC
execution  speed  has become Peter Norton's SYSINFO program.   The  version  of
SYSINFO  used for these illustrations was 3.0.  This program assigns a  numeric
rating to the computer tested, 1.0 being the speed of a standard PC running  an
8088 at 4.77 Mhz.  So, 1.0 is DEAD SLOW.

In  order  to  demonstrate just how much speed can be  gained  by  the  various
methods  detailed  above here are some operations, done with and  without  disk
cacheting  and  at  different clock speeds.  They were all  done  on  the  same
computer,  a  CHIPSET  (NEAT) '286 running at 10 or 20 Mhz.   This  machine  is
certainly not the fastest computer that money can buy, but it aint too shabby.

This machine consistently attains SYSINFO ratings of 11.5 or 15.5 (depending on
external  or internal bus timing) at it's 10 Mhz speed and 23.0 at it's 20  Mhz
speed.   There are utilities that will slow a computer down in order to play  a
video  game  designed  for  the  older,  slower  hardware.   Such  a   program:
VARISLOW.EXE was used to decrease the speed of the computer used for testing as
much as possible, to 3.6 or 4.0 SYSINFO rating.

Of  the many operations that can be done by DATAMAGE, sorting the records  into
order  is  the most demanding and time-consuming.  The sorts  were,  therefore,
used  for  the  benchmark tests.  The datafile used for the  bench  mark  tests
comprises  3,332 records. The size of this file after it was converted  to  the
DATAMAGE format (DATAMAGE also found and rejected all the duplicate records  it
contained during the conversion!) is 2,050,512 bytes.  It was imported from the
dBase  format, and is distributed with the SHAREWARE MARKETING SYSTEM from  Jim
Hood.  Thanks, Jim!

HARDWARE:

For  the hardware test the records were first arranged into an order  that  was
the same for each test, and completely random in terms of the desired order  on
RECORD  NUMBERS.  The record numbers are in the computer's memory, so the  disk
drive was not accessed.

SI RATE   1ST 1,000      2ND 1000       3RD 1000       TIME
===========================================================
3.6            1:15           3:20           5:25     12:11
15.5            :23           1:00           2:23      3:09
23.0            :14            :36            :58      2:11

As you can see, the few extra dollars spent to procure a modern, fast  computer
PAY OFF!  As I have learned, long ago: There's NO economy in buying junk!   And
there  is no point of reference that will demonstrate this axiom  more  clearly
and frequently than computer hardware.

HARDWARE and DISK CACHETING:

Even  more  difficult  than the sorting of numbers already  in  the  computer's
memory is the sorting of string data.  This must be gotten from the disk  drive
as the sort progresses, and takes far longer to compare than numeric data.

The same file contains a string field: COMPANY NAME.  The file was arranged  in
an  order that placed the company name at as near as you could come  to  random
intervals within the file, then sorted into alpha order.  As I watched the sort
progress I realized that each and every record was moved.

DBASE AND SORTING ALPHA:

While  completing this portion of the documentation I asked a young friend  who
has,  like  a lot of other people, been FAKED OUT by dBase, how long  it  would
take the industry standard to complete this task.  I pointed him to the file in
dBase format, and he got right back to me and, with swollen chest and in a VERY
loud voice, informed me that dBase did this in less than a minute.

I KNEW BETTER!

So,  I  went  over  there and asked him to do it for me.   He  gave  dBase  the
commands and it did SOMETHING in less than a minute.  What it did was to create
a whopping 355K b-tree file containing the data in the company name field.   It
did NOT sort the records nor, apparently, is it able to do so!

As  we  attempted  to USE this index we noticed that, while  getting  the  next
record  in  the  order, there was a delay.  This time was  spent  by  dBase  in
climbing it's b-tree in order to FIND said record.

DATAMAGE,  on the other hand, produced a MARKER file of only a little over  6K.
The  reading of the next record in the order was INSTANTANEOUS, as  the  MARKER
file  contained  it's  location  in the file.  DATAMAGE did  ALL  of  the  work
involved  in the sort at one time.  You could use the results of this work  for
the  next thousand years and not have to wait one additional nanosecond to  see
your records in the desired order.

In  the immortal words of one Laut Su (Chinese profit crica 1,000 A.D.):   "The
foremost  thoughts in a wise man's mind are these:  THINGS ARE SELDOM  AS  THEY
SEEM!"   Maybe  there is a program that will do a REAL alpha-sort  faster  than
DATAMAGE.   Be that as it may, I will NOT lie or cheat to make my program  SEEM
faster than it is.  But soon, maybe next year, I will re-write the BASE program
in C, and thereby honestly increase it's speed.

HONEST TIMES TO SORT THE DATA:

With  these times the speed of access to data on disk becomes a very  important
factor,  and disk cacheting begins to have great value.  The  proceeding  table
relates the computer's speed as well as the presence of cacheting.

NORTON     BYTES   1ST  RECORDS   2ND  RECORDS   3RD  RECORDS   TOTAL  RECORDS
SI RATE   CACHET  1000  SECOND   1000  SECOND   1000  SECOND     TIME  SECOND
==============================================================================
4.0            0 20:14     .78  29:53     .55  33:26     .50  1:37:14     .57
11.5           0 11:30    1.45  15:45    1.01  17:15     .97    50:13    1.10
11.5        1024  7:35    2.20   9:01    1.85  10:29    1.59    30:40    1.81
23.0           0  9:02    1.83  12.20    1.35  13:17    1.41    39:15    1.41
23.0        1024  5:08    3.25   5:29    3.03   7:23    2.26    18:07    3.07

You  are  quite  likely more mathematically reclined than  myself.   Hoards  of
numeric data could be extrapolated from the above - have fun!  Just one  point:
If  you  have a computer that actually rates 4.0 on SYSINFO you will  be  hard-
pressed  to match the first line.  Though the time wasting utility  slowed  the
microprocessor down this machine STILL communicates with a drive that's 3 times
as fast as the drive that came with the original PC or XT, down a 16-bit bus.

The point should be made, here, that DATAMAGE can, and usually will, sort  your
datafile  MUCH faster than this.  Look at the times above as sort of  a  worst-
case scenario.  See the file MAIN.DOC, under the heading F-10, MARKER FILE  and
the sub-heading RESTORE REMAINING RECORDS to discover how to get your work done
fast.   You will only need to sort the raw order ONCE, after that  you'll  make
much better time via the use of a MARKER FILE.

KEEPING THE FILES TOGETHER - FRAGMENTATION:

As  DATAMAGE  files  grow  your disk drive is bound  to  get  FRAGMENTED.   The
datafiles  will grow piece by piece.  You have a DATAMAGE file that is 100K  in
length; you enter several documents with your word processor, download a couple
hundred  Kbytes with your modem, then add some records to your  DATAMAGE  file.
All the things that happened in between will BE in between on your disk  drive,
resulting in DATAMAGE having to hunt all over the drive to read the file.

There  are utilities galore to defragment hard disks.  Two GOOD ones  that  are
available  as  SHAREWARE  from  your disk vendor or on  a  local  BBS  are  DOG
(DiskOrGanizer)  and  SST.  DOG is really safe, but also kinda  slow.   SST  is
really fast but, should your power fail during the session, bye bye disk!

Commercial  programs such as NORTON UTILITIES also include  defragmentation  in
their  bag  of tricks.  Whatever your choice of vehicle you  should  defragment
your  hard disk AT LEAST once a month, and for heavy usage once a week.   Doing
so will have you working more and waiting less.