:: Various Notes
-
tests performed on Dec. 3rd, 2019
-
comparing the original to this program
-
hypothesis: this program can produce visuals of higher resolution, with higher point count, with less memory per point
-
tested on /data/db/parse_db.0 (67.1 MB database file)
-
Banter 1.0 (Obj-C w/ OpenGL) stats
- program size: 556kb
- max points viewable: 262144
- RAM per
- 66k points (fixed): 110 MB
- CPU per
- 66k points (fixed): 30%
-
Banter 2.0 (C w/ OpenGL) stats
- program size: 42kb
- max points viewable: 500k (1 million w/ slowdown)
- RAM per
- 10k points: 18 MB
- 66k points: 25 MB
- 100k points: 29 MB
- 500k points: 72 MB
- 1mil points: 95 MB
- CPU per
- 10k points: 14% (6.1% stationary)
- 66k points: 24% (14.5% stationary)
- 100k points: 30%
- 500k points: 68%
- 1mil points: 88%
-
Results
- program 2.0 is 1400% smaller than the original program size
- program 2.0 can render 200%-400% more points
- program 2.0 has a 340% smaller memory footprint than program 1.0 (for 66k points)
- program 2.0 has a 25% smaller CPU footprint than program 1.0 (for 66k points)
-
Conclusions
- program 2.0 is superior in all measurements (memory footprint, # of points, program size, % cpu usage)
- program 2.0 has a much higher CPU % utilization as expected, the reduction over program 1.0 is only slight
- it appears program 2.0 is heavily CPU bound, similar to program 1.0 (as expected)
- anything that will reduce the CPU usage will result in significant increases in practical analysis limitations