If you only had these 4 properties this is very much the same as giving them indexes 1, 2, 3 and 4, as PofReader simply reads the next property index from the stream and checks if that’s the one you want to read, which is the case in vast majority of the cases.

That said, your conclusion still stands — there is a bit of extra work PofValue has to do to find the offset and length of each subsequent property in the stream, so accessing fourth property is more expensive than accessing third, which is more expensive than second, etc.

However, once you access property N, offsets and lengths of all properties between 0 and N are cached, so subsequent accesses to any of them has the same cost (it is a direct buffer access at that point). This means that accessing the last property first will make future access to any property cost the same (and be quite cheap, as stream parsing is what consumes most time), which might have interesting implications on criteria ordering in composite filters.

It would also be possible to optimize read performance (at expense of storage size) by pre-parsing the stream on entry update and storing property offset/length map as a binary decoration, but I’ll leave that as an exercise for your next blog post ;-) (keep them coming, btw, there is a lot of great stuff here)

Cheers,
Aleks

The main reason we added ASM was to implement partial classes for Coherence REST, but you can use it for annotation scanning and anything else as well (I just used it on a project to generate POF serialization code based on annotations)

Cheers,
Aleks