[Neo4j] Big Traverser perfomance : are '23 rels per milliseconds' good ?
michael.hunger at neotechnology.com
Thu Oct 6 21:25:21 CEST 2011
How much memory does your machine have?
What kind of disk is in there?
Have you looked at the memory config for the neo4j db?
What kind of scheduler do you use (please try deadline or as)?
Can you please share the config and JVM info that is output at the head of graphdb/messages.log ?
I'll attach some things we do to improve performance:
• Linux: scheduler used (cat /sys/block/*/queue/scheduler)
• Output from bonnie++ or other disk performance tool
• Please also run the write-test suite from here: http://wiki.neo4j.org/content/Linux_Performance_Guide
• Disk space, Used Swap
• ulimit -a (most probably open files should be >> 10k)
sysctl fs.file-max http://www.cyberciti.biz/faq/linux-increase-the-maximum-number-of-open-files
• encrypted fs hurts df -T /path/to/database/directory
Am 06.10.2011 um 21:03 schrieb Guillaume ALLEE:
> Hi all,
> I am using neo4j 1.5 java embedded. My traverser is a main java program
> http://pastebin.com/1ynVESbc which takes a db path as input.
> What it does is pretty basic : it follows my graph (which is a tree) and
> stores the ending leaves in a file. However it is quite big : I goes through
> 800.000 relationships and outputs 590.000 nodes in a txt file (11 MB)
> The global properties of my DB are the following :
> 962.008 nodes, 5.697.359 properties, 4.301.267 relationships, 2 relationship
> I do have the best results with the following JVM parameters : "-server
> -Xmx2500m -d32 -XX:+UseNUMA"
> JVM is 32 bits SUN JVM 1.6.0_24
> CPU is single core Intel(R) Xeon(R) CPU X5660 @ 2.80GHz
> OS: RHEL 5.5.
> As you can see in my code I did warm up the cache by making 2 traversers.
> First pass : 44 seconds
> 2nd pass : 34.5 seconds
> which make 23 relationships traversed per milliseconds.
> What do you think of that figures ?
> Do you see obvious things I should do to improve the speed of the traversal
> View this message in context: http://neo4j-community-discussions.438527.n3.nabble.com/Big-Traverser-perfomance-are-23-rels-per-milliseconds-good-tp3400711p3400711.html
> Sent from the Neo4j Community Discussions mailing list archive at Nabble.com.
> Neo4j mailing list
> User at lists.neo4j.org
More information about the User