On Mon, Sep 21, 2009 at 03:37:02PM -0400, Pavel Ivanov wrote: > > Have you any IO operations? As result you have dependence of page > > size. > > Though your performance most probably will not depend on these > operations because they will be executed at some random times by OS. > And they will be collected to have multiple blocks in one operation > anyway... > I don't have good knowledge of how disk cache works in kernel to say > if it will be beneficiary to send data there in chunks equal to blocks > on disk as opposed to chunks of any arbitrary size...
It is well-known that matching DB page size and filesystem record size is important for improving performance. SQLite3 supports power-of-2 page sizes from 512 to 32KB. Most filesystems use 512, 4096 or 8192 byte blocks. SQLite3's default pagesize is 1024 bytes. In the case of ZFS the fs recordsize is variable up to 128KB, and can be tuned per-dataset. You should set the recordsize to 32KB for ZFS datasets containing SQLite3 DBs, and you should set the SQLite3 pagesize to 32KB. Nico -- _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
