Re: findColumn performance

On Oct 6, 2008, at 6:36 PM, Alex Burgel wrote:

> ResultSetImpl.findColumn(String) uses a TreeMap with a comparator to
> find the correct column number.
>
> For big result sets with a lot of columns, this can be a bit of a
> bottleneck because the comparator must lower case the column names in
> the map for each lookup.
>
> I have been profiling a very db intensive portion of my app and I see
> findColumn as a hotspot. I use hibernate which gets columns by name,
> rather than index. I know thats not the most efficient way.
>
> For these cases, i was thinking it will be much faster to lower case  
> all
> the column names before putting them in the map. and then lower case  
> the
> key before you do a lookup on the map. you can even use a hashmap in
> this case which will give you a o(1) lookup time instead of o(log n).
>
> there will be a greater up front cost, tho i think that will be
> negligble.
>
> has anyone looked into this? i wanted to put this out there before i
> attempted a patch.
>
> --alex

Alex,

Alex, I see you've filed a bug on this, and your proposed patch (which  
I'm putting into 5.1.7) or fronting the lookup with another cache is a  
better approach, because it allocates less short-lived instances. The  
problem with .toLower() is that it creates a new String when it's  
called, and given that it will be called once for every column, for  
every row in the result set, that tends to get expensive as well.

	-Mark