Skip to main content

Ralsina.Me — Roberto Alsina's website

On how I accidentally may have made a feature 75% faster.

UP­DATE: Af­ter fix­ing bugs, it turns out the im­prove­ment is rough­ly 25%, not 75%, which is still nice :-)

Im­age gal­leries is one of my favourite fea­tures in Niko­la

I use them in my site and they are awe­some! Just dump a bunch of im­ages in a fold­er and you will have them nice­ly pre­sent­ed. Then you can add a file with a fold­er de­scrip­tion amd so on.

One thing I did not like was that they were pret­ty slow. My site has 3290 im­ages in the gal­leries sec­tion, and I dread­ed pro­cess­ing them be­cause they took for­ev­er, where for­ev­er means around 5 min­utes.

Un­re­lat­ed to per­for­mance, we had a fea­ture re­quest for sup­port­ing mul­ti­ple thumb­nail sizes, which is use­ful for things like pre­sent­ing the ide­al im­age size for your dis­play and such.

Well, to support that someday, a good first step is to have our generic image processor class support more than one thumbnail size. So, just take the scale_image implementation and have it support more than one size and more than one destination path, then loop over those, and that's that. Right?

OTO­H, it turns out we pro­cessed each im­age TWICE be­cause we would clean EXIF da­ta and/or re­size the "o­rig­i­nal" im­ages. Some­times the source im­ages have res­o­lu­tions that sim­ply make no sense on a web­site!

So, since our implementation of scale_image only accepted one destination, we would:

  1. read orig­i­nal im­age -> cleanup -> re­size to thumb size -> save
  2. read orig­i­nal im­age -> cleanup -> re­size to "large" size -> save

So, since I had now a ver­sion that could do the "re­size" and "save" parts on one cal­l, this be­came:

  1. read orig­i­nal im­age -> cleanup
  2. re­size to thumb size -> save
  3. re­size to "large" size -> save

And the re­al­ly slow part is ... read­ing the orig­i­nal im­age. So this sud­den­ly made the whole process take 25% of the time it took be­fore.

In my spe­cif­ic site, it went down from 356 sec­onds to 112 sec­ond­s. That's a 70% de­crease in ren­der­ing time. Which is big. And in code I first wrote around 2013, that is huge.

So, the les­son is ... I am a crap­py pro­gram­mer, may­be? Or used to be? Or maybe just that even old, ma­ture code could still have large op­por­tu­ni­ties for im­prove­men­t?

Who knows.