On 1/26/2013 12:55 AM, Rahul Bishnoi wrote:
Thanks for quick reply and addressing each point queried.
Additional asked information is mentioned below:
OS = Ubuntu 12.04 (64 bit)
Sun Java 7 (64 bit)
Total RAM = 8GB
SolrConfig.xml is available at http://pastebin.com/SEFxkw2R
Rahul,
The MaxPermGenSize could be a contributing factor. The documents where
you have 1000 words are somewhat large, though your overall index size
is pretty small. I would try removing the MaxPermGenSize option and see
what happens. You can also try reducing the ramBufferSizeMB in
solrconfig.xml. The default in previous versions of Solr was 32, which
is big enough for most things, unless you are indexing HUGE documents
like entire books.
It looks like you have the cache sizes under <query> at values close to
default. I wouldn't decrease the documentCache any - in fact an
increase might be a good thing there. As for the others, you could
probably reduce them. The filterCache size I would start at 64 or 128.
Watch your cache hitratios to see whether the changes make things
remarkably worse.
If that doesn't help, try increasing the -Xmx option - first 3072m, next
4096m. You could go as high as 6GB and not run into any OS cache
problems with your small index size, though you might run into long GC
pauses.
Indexing, especially big documents, is fairly memory intensive. Some
queries can be memory intensive as well, especially those using facets
or a lot of clauses.
Under normal operation, I could probably get away with a 3GB heap size,
but I have it at 8GB because otherwise a full reindex (full-import from
mysql) runs into OOM errors.
Thanks,
Shawn