Re: Slow Group-By

[mamaco <ma...@163.com>]

hi mike,
you can create another cache with key equals (idnumber + value), redirect
your streamer to it, and then feel free to check cache size. it could waste
your RAM, but fast enough for any parallel queries, because it's not
map/reduce at all.  we used this weird k/v trick in many 'ignite reports'. 

marco




Williams, Michael wrote
> All,
> 
> Any advice on speeding up group-by? I'm getting great performance before
> the group-by clause (on a fairly decent size data set) , but adding it
> slows things down horribly. Without the group-by clause, the query takes
> about a minute, and all cores are fully in use. With the group-by cores,
> the query takes far longer. The only difference in the EXPLAIN plans
> appears to be the addition of the group-by clause.  There are a lot of
> rows to be joined in the clause. Is there an equivalent to
> setQueryParallelism for this? Any things to try?
> 
> Thanks,
> Mike





--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/