Using the formula I supplied in my previous reply I'd say you can go
with a 32G heap or just slightly lower.

2011/2/24 Pablo Pareja <[email protected]>:
> Hi Mattias,
>
> I guess that my only option would then be limiting the max amount to 2GB as
> you say,
> however this will make the insertion phase way longer than what I expected
> and counted on.
> Just to confirm, am I then supposed to launch my java process with at most
> -Xmx2G ??
>
> I really think this is an important flaw to work on.
> I look forward to seeing it fixed in the near future.
>
> Cheers,
>
> Pablo
>
>
> On Thu, Feb 24, 2011 at 3:19 PM, Mattias Persson
> <[email protected]>wrote:
>
>> There's this hardcoded setting for batch inserter indexes that
>> increases the ram buffer size to increase overall performance during
>> batch insertion. The formula is:
>>
>>        double atLeast = writer.getRAMBufferSizeMB();
>>        double heapHint =
>> (double)(Runtime.getRuntime().maxMemory()/(1024*1024*16));
>>        double ramBufferSize = Math.max( atLeast, heapHint );
>>        writer.getRAMBufferSizeMB( ramBufferSize );
>>
>> It's just something which I ballparked and obviously has problems for
>> heaps of your size. It's a tough one to fix in your case since you
>> cannot parameterize that calculation. The only way would be to modify
>> source code and build a new jar, or wait for it to be fixed (maybe by
>> limiting the max amount to 2Gb as Lucene seem to suggest). Hopefully
>> it can be addressed soon.
>>
>> 2011/2/24 Pablo Pareja <[email protected]>:
>> > Hi all,
>> >
>> > I just got this exception:
>> >
>> > ramBufferSize 3584.0 is too large; should be comfortably less than 2048
>> > SEVERE:
>> >
>> org.apache.lucene.index.IndexWriter.setRAMBufferSizeMB(IndexWriter.java:1368)
>> > SEVERE:
>> >
>> org.neo4j.index.impl.lucene.LuceneBatchInserterIndex.instantiateWriter(LuceneBatchInserterIndex.java:209)
>> > SEVERE:
>> >
>> org.neo4j.index.impl.lucene.LuceneBatchInserterIndex.<init>(LuceneBatchInserterIndex.java:66)
>> > SEVERE:
>> >
>> org.neo4j.index.impl.lucene.LuceneBatchInserterIndexProvider.index(LuceneBatchInserterIndexProvider.java:135)
>> > SEVERE:
>> >
>> org.neo4j.index.impl.lucene.LuceneBatchInserterIndexProvider.nodeIndex(LuceneBatchInserterIndexProvider.java:102)
>> > SEVERE:
>> >
>> com.era7.bioinfo.bio4j.programs.ImportGeneOntology.main(ImportGeneOntology.java:107)
>> > SEVERE:
>> >
>> com.era7.bioinfo.bio4j.programs.ImportGeneOntology.execute(ImportGeneOntology.java:75)
>> > SEVERE:
>> >
>> com.era7.lib.bioinfo.bioinfoutil.ExecuteFromFile.main(ExecuteFromFile.java:66)
>> > SEVERE:
>> >
>> com.era7.bioinfo.bio4j.programs.CreateWholeBio4j.main(CreateWholeBio4j.java:28)
>> >
>> > when launching my Batch insertion program once I had updated the
>> indexation
>> > to the new API
>> > any ideas about what should I do?
>> >
>> > I'm launching the program in a 68GB RAM machine with the following
>> > batchInserter.properties:
>> >
>> > neostore.nodestore.db.mapped_memory=10G
>> > neostore.relationshipstore.db.mapped_memory=12G
>> > neostore.propertystore.db.mapped_memory=12G
>> > neostore.propertystore.db.strings.mapped_memory=10G
>> > neostore.propertystore.db.arrays.mapped_memory=512M
>> >
>> > I don't know how or where to change that "ramBufferSize' value, which by
>> the
>> > way didn't specified at any point.
>> >
>> > Thanks in advance,
>> >
>> > Pablo
>> >
>> >
>> > --
>> > Pablo Pareja Tobes
>> > LinkedIn    http://www.linkedin.com/in/pabloparejatobes
>> > Twitter       http://www.twitter.com/pablopareja
>> >
>> > http://about.me/pablopareja
>> > http://www.ohnosequences.com
>> > _______________________________________________
>> > Neo4j mailing list
>> > [email protected]
>> > https://lists.neo4j.org/mailman/listinfo/user
>> >
>>
>>
>>
>> --
>> Mattias Persson, [[email protected]]
>> Hacker, Neo Technology
>> www.neotechnology.com
>> _______________________________________________
>> Neo4j mailing list
>> [email protected]
>> https://lists.neo4j.org/mailman/listinfo/user
>>
>
>
>
> --
> Pablo Pareja Tobes
> LinkedIn    http://www.linkedin.com/in/pabloparejatobes
> Twitter       http://www.twitter.com/pablopareja
>
> http://about.me/pablopareja
> http://www.ohnosequences.com
> _______________________________________________
> Neo4j mailing list
> [email protected]
> https://lists.neo4j.org/mailman/listinfo/user
>



-- 
Mattias Persson, [[email protected]]
Hacker, Neo Technology
www.neotechnology.com
_______________________________________________
Neo4j mailing list
[email protected]
https://lists.neo4j.org/mailman/listinfo/user

Reply via email to