T4 DIB Library Benchmark Data

This benchmark illustrates the benefits of consolidating
individual image-files into a compressed DIB Library.

The machine used was a 486DX-33, 8MB RAM, Windows 3.1 with
SmartDrv set at 1MB. The video was 256-color 1024x768.

The images chosen were 106 256-color TIF files. Each file was
about 25K. The aggregate size of the files was 26MB. I then
created a DIB Library from these members (all 106), compressing
all of them. The resulting DIB Library size was 14MB.

SPACE SAVINGS: The DIB Library was 54% of the size of the
original total file sizes. In general, 256-color images compress
40-50%, and 16-color images compress 85%+.

***********

To determine performance, I first sequentially displayed and
timed individual TIF files (load and show), and then did the same
for the same images from a fully compressed DIB Library. I used a
variety of storage media, specifically:

- Hard-drive (EISA SCSI, pretty fast), all 106 images. I used
  both a standard-spaced and a double-spaced volume.
- IOMEGA Bernoulli 90, all 106 images.
- NEC 84 CD-ROM drive, all 106 images, TIF files only.
- 3-1/2" Diskette drive, 5 images.

For all of these tests, the volumes were read-cached to a 1MB
SmartDrive.

The following shows the frame-per-second (fps) performance of
these tests. Note that since I had no way of writing a DIB
Library to the CD-ROM drive, I extrapolated that value based on a
straight-line projection between the two other closest media
(Diskette and Bernoulli). This projection is probably not
completely accurate, but it's better than nothing.

              FRAMES PER SECOND
                RAW     DIB      PERCENT
MEDIA          FILES  LIBRARY  IMPROVEMENT
Std Spaced HD  .475    .434      - 9.7%
DblSpaced HD   .427    .445        4.2%
Bernoulli      .391    .400        2.3%
CD-ROM         .228    .259 *     13.6% *
Diskette       .063    .082       30.2%

* = Extrapolated based on straight-line approximation.

In general, performance of a compressed DIB Library, compared to
single-image files, INCREASES as:
- CPU performance characteristics INCREASE and/or
- Media performance characteristics DECREASE and/or
- SmartDrive cache-hit ratios DECREASE.

Since SmartDrv was a "wildcard" in the above benchmark
(unpredictable cache hit/miss), I ran a duplicate demo, but with
SmartDrv shut off altogether. I know none of your users would
think of working this way, but this next benchmark does show the
effect of a 0% cache-hit situation. Again, I did an approximation
for the effect of a library on the CD-ROM.

              FRAMES PER SECOND
                RAW     DIB      PERCENT
MEDIA          FILES  LIBRARY  IMPROVEMENT
Std Spaced HD  .393    .438       11.5%
DblSpaced HD   .402    .445       10.7%
Bernoulli      .344    .408       18.6%
CD-ROM         .230    .286 *     24.3% *
Diskette       .049    .094       91.8%

* = Extrapolated based on straight-line approximation.

The above benchmark can be used to show another dimension - how
much the performance VARIES with SmartDrv on and off. These
values can be used to illustrate CONSISTENCY in performance. I
have left out the CD-ROM in this table, since that data is
extrapolated anyway. 

             PERFORMANCE VARIANCE
MEDIA           FILES   LIBRARY
Std Spaced HD   20.9%    1.0% 
DblSpaced HD     6.2%    0.0%
Bernoulli       13.7%    1.6%
Diskette        28.6%   12.8%

As you can see, the DIB Library is less sensitive to SmartDrv
issues than the single-image files.

Finally, just because someone might want to see it, I included a
benchmark on uncompressed libraries. You may find the results
somewhat surprising in that uncompressed members do not always
significantly outperform compressed members. This is because the
time to read the member vs. the time to decompress it is very
dependent upon a specific machine configuration.

              FRAMES PER SECOND
                RAW     DIB      PERCENT
MEDIA          FILES  LIBRARY  IMPROVEMENT
Std Spaced HD  .475    .616       29.7%
DblSpaced HD   .427    .535       25.3%
Bernoulli      .391    .471       20.5%
CD-ROM         .228    .268 *     17.5% *
Diskette       .063    .060      - 1.6%

* = Extrapolated based on straight-line approximation.

From this, you can see the DIB Library can, under the right
circumstances, offer a significant performance improvement over
single-image files. Since you can elect compression on a per-
member basis, if you know the kind of device your library will
wind up on, you can tune the library for the best combination of
performance and size.