Re: Solr in a highly memory constrained environment (ie. VPS) - stupid idea?

[Yonik Seeley <yo...@apache.org>]

Yup, should definitely be doable with the small number of docs and traffic.
- comment out the query cache in solrconfig.xml
- reduce the size of the document cache to maybe 50 entries or so
- reduce maxBufferedDocs to 10 to limit indexing memory (a better option will be
  available soon in Solr 1.3-dev)
- if you are using faceting, try setting facet.enum.cache.minDf=10000
to avoid using the filterCache altogether

-Yonik


On Mon, Mar 3, 2008 at 1:04 PM, Micah Wedemeyer <mw...@emory.edu> wrote:
>  I've used Solr a little at work where we have our own hardware with all
>  the memory we want.  However, I would also like to use Solr on a
>  small-ish website that I run off of a VPS with 512MB of RAM.  I tried
>  this (untuned) for a while, and Tomcat/Solr would just grab up all my
>  memory until the server slowed to a crawl.
>
>  Performance really isn't a huge factor.  There are probably around 5000
>  documents with an average size of 1-2k per document.  A document (wiki
>  page) is re-indexed roughly every time it is saved, and with our
>  current traffic, that's probably around 5/minute at our busiest times.
>  Plus, the search feature isn't all that prominent, meaning that actual
>  searches occur fairly rarely, perhaps 1/minute at peak.
>
>  So, my basic question is: Can I constrain and constrict Solr in order
>  to run it in a tiny slice of memory, say 64-128MB?  What will happen if
>  I set the -xmx option to 64M?  I saw some other posts about OutOfMemory
>  errors, but I couldn't get a clear idea about how well Solr recovered
>  from them.

>  Also, is there a way to turn off (or severely limit) caching and
>  buffering?  What happens if I set maxBufferedDocs to 1 or 10?  Can I go
>  through and set the filter, query, and document caches to some small
>  number like 1 or 10?
>
>  Finally, am I trying to do something that's just plain stupid?  I like
>  Solr and I'd like to try some neat stuff with facets on my site, but
>  Sphinx ( http://www.sphinxsearch.com/ ) is another option that I know
>  people have gotten to work with a small memory footprint.  I just don't
>  know Sphinx that well.
>
>  Thanks in advance,
>  Micah
>
>