On 2016/03/06 10:00 AM, Keith Medcalf wrote: > On Saturday, 5 March, 2016 12:12, Richard Hipp said: > >> A key point of this email thread is that SQLite performance seems to >> be maximized when the database page size matches the CPU memory page >> size and the disk sector size - currently 4096 bytes on most hardware. > This does not seem to be entirely true, at least not on Windows 10 1511 Pro > with a running on a very fast SSD. > I suspect it is more related to the size and shape of the data being stored > and the operations being performed. > Clearly the best thing that can be done to improve performance is increase > the internal pagecache. > > These are benchmark times for --size 500 on Windows 10 ... > > Time= 66.81, Cachesize= 65536, Pagesize= 8192, Cacheram=536870912 > Time= 67.10, Cachesize= 32768, Pagesize= 8192, Cacheram=268435456 > Time= 67.37, Cachesize= 4096, Pagesize= 8192, Cacheram=33554432 > Time= 67.92, Cachesize= 8192, Pagesize= 8192, Cacheram=67108864 > Time= 68.00, Cachesize= 16384, Pagesize= 8192, Cacheram=134217728 > Time= 68.31, Cachesize= 131072, Pagesize= 8192, Cacheram=1073741824 etc..
Fascinating stuff, thanks Keith - mind posting the DB or schema and the query(ies) used during the benchmark tests? Seems from here the 8k pagesize with 64k cache makes sense, unless that schema turns out to be wildly non-standard. Thanks, Ryan
