lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael McCandless <luc...@mikemccandless.com>
Subject Re: Understanding the IndexWriter-Infostream log
Date Thu, 17 Mar 2011 20:32:48 GMT
Hi Tom,

Answers below...

On Thu, Mar 17, 2011 at 1:19 PM, Burton-West, Tom <tburtonw@umich.edu> wrote:
> Hello all,
>
> We have very large documents with large numbers of unique terms.   Our documents average
about 800,000 KB and about 200,000 tokens.  In trying to understand how often the ramBuffer
gets flushed to disk we turned on the IndexWriter log.
> <infoStream file="/tmp/IndexWriter.log">true</infoStream>
>
> With the Solr default setting of ramBufferSizeMB=32 it appears that the buffer is flushing
every 2-10 documents.
>
> When we see this in the IndexWriter log: "flush postings as segment _9hw numDocs=2",
 does this mean that the buffer is writing a segment that contains only 2 documents?

Yes.

> Also how do we interpret the following?
>
> "DW:   ramUsed=33.467 MB newFlushedSize=6764406 docs/MB=0.155 new/old=19.276%"
>
> What does "docs/MB" and 'new/old" mean?

docs/MB is docCount divided by size of the newly flushed segment's size in MB.

new/old measures "RAM efficiency", ie size of the flushed segment
divided by RAM consumed before the flush.  In this case IndexWriter
flushed because the RAM buffer hit 33.467 MB, but then the newly
flushed segment was 6.45 MB (= 6764406 / 1024 / 1024).  This number is
expected to be well below 100%, ie, we pay a price to have the
malleable data structures in RAM that allow for incoming docs to be
inverted... though a "good" number is maybe 30%.  The more unique
terms you have the lower this number will be...

-- 
Mike

http://blog.mikemccandless.com

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


Mime
View raw message