[Xapian-discuss] Initial benchmark results quartz and flint

Arjen van der Meijden acmmailing at tweakers.net
Thu Jun 30 21:28:25 BST 2005


I've now done all three runs. And I verified that the query results from 
the non-compacted quartz-database are exactly the same as for the 
non-compacted flint-database. Would you like me to verify that the 
compacted databases yield the same results as well?

Here are the benchmark runtimes for the various databases:
Non-phrase			
Flint             155,574  160,540  161,227
Flint compact      96,569   95,676   97,584
Flint compact -F   94,227   94,620   95,028

Quartz            169,853  161,802  161,397
Quartz cpt -F gz  108,783  110,264  108,911
Quartz cpt -nF gz 109,650  108,199  108,062
Quartz cpt        103,863  103,676  103,718
Quartz cpt 084 gz 108,299  108,221  109,241
			
Phrase			
Flint             2841,680  2826,288  3152,847
Flint compact     3026,939  2582,435  2924,166
Flint compact -F  2623,404  2643,625  2608,039

Quartz            7037,056  6612,721  6407,643
Quartz cpt -F gz  9249,504  8032,758  9598,598
Quartz cpt -nF gz 8090,707  8038,627  8153,904
Quartz cpt        9410,721  8389,753  8059,687
Quartz cpt 084 gz 8100,171  8124,942  8381,013

For the test-setup, see my previous e-mail.

I'm still very puzzled with the results of the various compacted quartz 
databases with phrase queries. I'm also a bit clueless as to the huge 
differences in times anyway, eventhough the machine was as far as I know 
hardly in use. And the disk, with scsi-characteristics, was dedicated to 
the benchmark.

Anyway, I think the fuller-compacted flint is a clear winner in 
performance. But another conclusion that can be drawn is that trying to 
decrease the database size won't always help in performance.  Atleast 
the results seem to indicate that applying zlib to the smaller tables 
tends to do more worse than good.
The test machine has much less memory than our production machine, so in 
our production it'll be even less I/O-bound. But I have no real clue how 
that would translate in performance with these results.

Best regards,

Arjen



More information about the Xapian-discuss mailing list