4 ms·

This is an interesting write-up although I'd question the take home at the end: "It really doesn’t take much to find 10x wins." Sometimes there's a really obvi

This is an interesting write-up although I'd question the take home at the end: "It really doesn’t take much to find 10x wins." Sometimes there's a really obvious problem that you can solve that will get you that kind of 10x performance boost. Often though you have to work your ass off for it or, even having found a problem where fixing it might even double your performance, the fix might be complex, touching multiple parts of the system. In my experience, whilst you can often find quick wins, and sometimes these will be enough, there will always be cases where the analysis is much harder, or where more complex fixes are required. (Disclaimer: I am a performance consultant, focussed mainly on SQL Server and .NET. My background is that years and years ago I lead the team that built Red Gate's ANTS Performance Profiler and ANTS Memory Profiler, and also worked on a number of their SQL Server tools.)

jng 9y ago
This. I would call the optimization described on the article as "low-hanging fruit": if you don't have the right DB indices or you are not batching your operations, you can find that kind of improvement easily, mostly just because your code was not following the "fast path". Still very useful article even if just to show how to use profiling tools, which you will need to even for much-higher-hanging fruit.
mason55 9y ago
Very similar to Ben Horowitz's "Lead Bullets" story: http://a16z.com/2011/11/13/lead-bullets/ http://a16z.com/2011/11/13/lead-bullets/
mpweiher 9y ago
> I'd question the take home at the end: "It really doesn’t take much to find 10x wins." Hmm. I guess it depends on what you mean with "much". In terms of how capable our hardware is and how comparatively slow our software, yes, it really doesn't take much, there are typically orders of magnitude to be had. Sometimes that means ditching assumptions, however, and that can be hard. For example, I've often had pretty easy 10x, 100x and 1000x+ wins by removing SQL databases and associated ORMs.

jng 9y ago

This. I would call the optimization described on the article as "low-hanging fruit": if you don't have the right DB indices or you are not batching your operations, you can find that kind of improvement easily, mostly just because your code was not following the "fast path". Still very useful article even if just to show how to use profiling tools, which you will need to even for much-higher-hanging fruit.

mason55 9y ago

Very similar to Ben Horowitz's "Lead Bullets" story: http://a16z.com/2011/11/13/lead-bullets/ http://a16z.com/2011/11/13/lead-bullets/

mpweiher 9y ago

> I'd question the take home at the end: "It really doesn’t take much to find 10x wins." Hmm. I guess it depends on what you mean with "much". In terms of how capable our hardware is and how comparatively slow our software, yes, it really doesn't take much, there are typically orders of magnitude to be had. Sometimes that means ditching assumptions, however, and that can be hard. For example, I've often had pretty easy 10x, 100x and 1000x+ wins by removing SQL databases and associated ORMs.