bands slow, especially with filters

Case number:845813-2008392
Topic:Game: Tools
Opened by:LociOiling
Status:Open
Type:Suggestion
Opened on:Monday, December 16, 2019 - 02:58
Last modified:Tuesday, December 17, 2019 - 23:42

As discussed recently, bands seem very slow. Filters slow things down even more.

The recipe bandit 1.0 demonstrates the issue. It adds a zero-length spaceband to each segment, while varying the unto and filter settings. The whole process repeats a given number of times. (Thanks to Paul Dunn for the base code used in this experiment.)

The table below shows results for puzzle 1773. The results will be easier to see if you copy the lines below and paste them into a spreadsheet.

The results show that leaving both undo and filters enabled is the slowest option, averaging over 48 seconds to add 90 bands.

Turning off both undo and filters is fastest, at just over 19 seconds.

Leaving undo enabled while disabling filters slows things to 22 seconds.

Disabling undo while leaving filters enabled is second slowest, at over 46 seconds.

case	disable undo	disable filters	total time	runs	mean time
1	false	false	1450	30	48.333333333333
2	true	true	571	30	19.033333333333
3	false	true	671	30	22.366666666667
4	true	false	1382	30	46.066666666667

On puzzle 1772, which has only a single "Residue Count" filter, the results aren't as clear, but in limited testing, there still seems to be slight advantage to disabling both undo and filters. (More results on 1772 later.)

(Mon, 12/16/2019 - 02:58  |  2 comments)


jeff101's picture
User offline. Last seen 8 hours 15 min ago. Offline
Joined: 04/20/2012
Groups: Go Science
The chart above is:

case    disable undo    disable filters    total time    runs    mean time
1	false	        false	           1450	         30	 48.333333333333
2	true	        true	            571	         30	 19.033333333333
3	false	        true	            671	         30	 22.366666666667
4	true	        false	           1382	         30	 46.066666666667
bkoep's picture
User offline. Last seen 2 days 7 hours ago. Offline
Joined: 11/15/2012
Groups: None

Thanks for benchmarking this, LociOiling! This is definitely a bug.

I've confirmed that there are some redundant filter calculations going on when a band is created, which would account this kind of slowdown. I haven't pinpointed the source of the bug yet (it looks like a cache is being invalidated somehow, which could be caused by a lot of things), but hopefully will have some more time to look at it this week!

Sitemap

Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, RosettaCommons