Comments on: Optimizing Laszlo Explorer: Laziness is good

By: Me

Me — Thu, 01 Jun 2006 13:08:14 +0000

Tom: lol.

By: Tom

Tom — Mon, 15 May 2006 12:50:16 +0000

I think to be fair 20 seconds is damn quick if you are infact running it on 1.6Mhz powerbook?

By: jgrandy

jgrandy — Fri, 21 Apr 2006 05:05:55 +0000

Simon - I’m not myself sure whether the purely recursive design of tree, or the flattened design of opttree, could stand alone given the underlying performance constraints we operate with. The original tree design can be made fast if it is completely lazy — but that requires specifying subnode classname and letting the tree instantiate its children on demand rather than using the declarative syntax that we like so much. On the other hand, opttree is very fast but isn’t as appropriate if the secondary levels of the tree have very different look and feel from the top levels.

I do think the original tree could be made to support dynamic data with a moderate amount of work; I just don’t think that was considered by the original designer.

By: Simon O'Connor

Simon O'Connor — Fri, 21 Apr 2006 02:00:46 +0000

Great breakdown of performance tips and a good insight into how some things work within Laszlo.

On your last note also, while you say that the component can be made to be effecient and is elegant, my issue with it has always been it’s handling of dynamic data, which is very poor. I have switched our application over to the new tree in the incubator of 3.2 and found it works wonderfully with dynamic data, and also seems quicker to run also.

What are your thoughts on its makeup in comparison. (As it directly alters the dataset its representing, as well as being a somewhat ‘flat’ design rather than recursive)

By: Raju

Raju — Thu, 20 Apr 2006 11:53:21 +0000

Hi Jim, excellent breakdown and explanation of the problem. One of the things I’m most afraid with OpenLazlo development is the unnecessary instantiation of a high number of views/subnodes on application startup. The performance with the Flash 8 plugin improved a lot, but still you have to carefully consider which parts of your applications have to be instantiated and ready for usage when the splash screen is removed.

The last note is interesting: I would have thought, too, that the recursive design of the basetree would be responsible for the slowness. But there are always surprises around when improving application performance. Looking forward to your post on the OpenLazslo profiler.