Hi,

If you use FSDirectory.open() it will automatically choose MMapDirectory on 64 
bit systems. Please note, virtual memory is != physical RAM. A 64 bit machine 
has *always* >1 Terabyte of virtual address space available, this is unrelated 
to physical memory (a common misunderstanding about mmap): 
http://blog.thetaphi.de/2012/07/use-lucenes-mmapdirectory-on-64bit.html

The speed difference depends on use case: In general MMapDirectory is much 
faster in multi-threaded environments, because no concurrency problems. If you 
use SimpleFSDirectory this is the largest bootleneck. NIOFSDirectory does not 
have concurrency problems, but it is still slower because it does a lot of 
extra copying of data between kernel space and user space for buffering. 
MMapDirectory is muuuuuuuch faster if you sort by docvalues fields, because it 
supports random access without any buffering overhead.

So please: Use MMapDirectory where possible - this is completely unrelated to 
how much RAM you have available!

Uwe

-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
http://www.thetaphi.de
eMail: u...@thetaphi.de


> -----Original Message-----
> From: sreedevi s [mailto:sreedevi.payik...@gmail.com]
> Sent: Thursday, February 05, 2015 10:13 AM
> To: java-user@lucene.apache.org
> Subject: MMapDirectory or FSDirectory
> 
> Hi,
> I am doing some performance analysis with lucene. I have 1 million resources
> with 1000 attributes.
> According to how I index, I will have 1 million documents with 1000 fields.
> For me the total data was about 100 GB and while using FSDirectory to store
> my indices, index size was almost 6 GB.
> I have virtual memory available of almost 8 GB. Is it advised to use
> MMapDirectory for increased performance?
> Many blogs suggest it doesnt bring out much performance difference.
> 
> 
> Best Regards,
> Sreedevi S


---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org
For additional commands, e-mail: java-user-h...@lucene.apache.org

Reply via email to