[erlang-questions] [clarify] mnesia slowdown with extra indexes ???

[Zoltan Peter Toth]

Hi,

I made some test on mnesia object matching on non-key attributes.

Got a table like this:
 mnesia:table_info(tgccDevIdToTerminationTable,size).
27000                              

 mnesia:table_info(tgccDevIdToTerminationTable,attribute.)                                                                  

[deviceId,terminationId,sessionId,mgId,deviceChar, deviceBitMask,chId]

No extra index exists.
A match on the mgId attrib gives:

element(1,timer:tc(mnesia,dirty_match_object,[tgccDevIdToTerminationTable, 
{'_','_','_','_',200021,'_','_','_'}])).
5578
5612
5571
6064
5622

Now I added an index on the mgId:
 mnesia:add_table_index(tgccDevIdToTerminationTable, 
mgId).                                                       
{atomic,ok}

...and the matches slowed down:
6542
6420
6302
6430
6519

This is determinitstic. If I remove the extra index, it becomes faster 
again.
The same even if specifying the index explicitly with 
dirty_index_match_object().
This is a live system with all the mnesia nodes up. R12B on quad core.

Any comment on this ? Or is a table with 27000 rows too little to take 
advantage of the
extra index ???
Br,
    TZP.