On 9/27/06, Ken Krugler <[EMAIL PROTECTED]> wrote:
One trick I've heard is to do cat <path to Lucene index directory>/_*.* > /dev/null to force all of the index data into the OS cache.
For Solr, I don't think I'd do this: - If you are using replication, the new segment files that just got rsync'd *will* be in memory already. - warming/autowarming can more intelligently exercise the parts of the index that need to be cached. For example, if the index is bigger than memory, you want the actual index files cached before any of the stored fields. So it could still make sense to do this on a cold boot if you actually have enough spare RAM to cache the whole index in memory. You should still have some warming in Solr though for things like loading the term index, the field norms, and any FieldCache entries that will be used for sorting. -Yonik
