[Mondrian] Performance: CellInfoPool

Julian Hyde julianhyde at speakeasy.net
Mon Aug 13 19:06:03 EDT 2007


This looks like a performance bug.

However, I ran some tests on ObjectPool using Integer and String values and
it looks competitive with HashSet in terms of speed. So I don't think
there's a performance bug in ObjectPool.

Can you confirm that this still occurs in 2.4? If so, please log a bug.

Julian

PS Here are the numbers. I added 300,000 entries to both an ObjectPool and a
HashSet, then did 700,000 retrieves. The density affected the number of
duplicate keys which were generated.

        // Key type:        Integer               String
        // Implementation:  ObjectPool    HashSet ObjectPool    HashSet
        //                  ========== ========== ========== ==========
        // With density=0.01, 298,477 distinct entries, 7,068 hits
        // 300,000 adds        221 ms      252 ms     293 ms    1013 ms 
        // 700,000 gets        164 ms      148 ms     224 ms     746 ms
        //
        // With density=0.5, 236,022 distinct entries, 275,117 hits
        // 300,000 adds        175 ms      250 ms     116 ms     596 ms
        // 700,000 gets        147 ms      176 ms     190 ms     757 ms
        //
        // With density=0.999, 189,850 distinct entries, 442,618 hits
        // 300,000 adds        128 ms      185 ms      99 ms     614 ms
        // 700,000 gets        133 ms      184 ms     130 ms     830 ms


> -----Original Message-----
> From: mondrian-bounces at pentaho.org 
> [mailto:mondrian-bounces at pentaho.org] On Behalf Of Anton Nikitin
> Sent: Monday, August 13, 2007 9:19 AM
> To: 'Mondrian developer mailing list'
> Subject: [Mondrian] Performance: CellInfoPool
> 
> 
> Hi All.
> 
> It seems to me that CellInfoPool is too much slow on huge queries.
> 
> I use customized Mondrian version based on 2.3.2
> 
> When running a query which results in 300000 cells I'm observing that
> CellInfoPool implementation is 10 times slower than CellInfoMap.
> 
> In my case I simply turned off CellInfoPool in all case 
> (memory is less
> significant for us than performance). 
> 
> Anton
> 
> _______________________________________________
> Mondrian mailing list
> Mondrian at pentaho.org
> http://lists.pentaho.org/mailman/listinfo/mondrian
> 




More information about the Mondrian mailing list