lucenenet-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "George Aroush" <>
Subject RE: Performance of Dot Lucene
Date Thu, 17 Aug 2006 16:13:42 GMT
Thanks Neil.  Let us know what you find.


-- George 

-----Original Message-----
From: Neil Carson [] 
Sent: Thursday, August 17, 2006 11:23 AM
Cc: Mark Brinicombe
Subject: RE: Performance of Dot Lucene

Hi George
I CC'd Mark Brinicombe here who ran the tests. We'll see if he can give it a
Maybe the IKVM'd version is somehow using nio...


From: George Aroush []
Sent: Thu 8/17/2006 7:38 AM
Subject: RE: Performance of Dot Lucene

Hi Neil,

Since you have the environment to test performance, can you try one more
test?  If so, please run a test for Lucene 1.4.3 -- both the Java version
and the C#.  Why?  With Lucene 1.9 and up, Lucene started using "java.nio.*"
(see the file Lucene.Net.Store.MMapDirectory.cs)  This code is much faster
then it's non-NIO version.  Unfortunately, in the C# version, I have not
been able to port this from Java to C#.

All: Anyone want to take on porting this file, please let me know.


-- George Aroush

-----Original Message-----
From: Neil Carson []
Sent: Wednesday, August 09, 2006 7:37 PM
Subject: Performance of Dot Lucene

Hi all

Was curious about any performance measurements folks have done with the .NET
port of Lucene.

I did some simple measurements, and the results are repeatable (each test
ran three times, each run within a second or two of the previous) and quite
surprising. The test case in question was indexing a 700MB text document
with all of Lucene 1.9's default options, using optimised Release builds.
Elapsed times are  min:sec:

Java Lucene -server VM:        3:29
Java Lucene -client VM:        4:02
Java Lucene/IKVM'd on .NET 2:  4:40
DotLucene, .NET 2:             5:03
Java Lucene ported/J#, .NET 2: 6.20

I didn't do any profiling (we will get to this later, when we do our Lucene
implementation and need to start performance tuning); however the results
were quite surprising. From the task manager, in all cases the average was
probably 50% kernel time and 50% user time. In all cases, all files were
reset, caches flushed, no other procs running etc.

Has anyone else ran any similar tests, especially with querying? Could this
be down to the quality of the Java compiler vs C#, or the respective VMs -


View raw message