Aris Setyawan wrote:
SQLightning replaces the SQLite backend with Symas' LMDB, which also uses
MVCC
and thus supports high concurrency. It is also many times faster than
BerkeleyDB and vanilla SQLite.
Your MVCC is different compared to InnoDB or BDB locking. Every one
should carefully read each DB's doc, then test it before decide to use
it.
Yes, it's different. In LMDB writers never block readers and readers never
block writers. The original poster was complaining about SELECT taking a long
time and preventing other threads from making progress. That problem doesn't
exist in LMDB. BDB locking *might* be able to avoid this in many cases, if
there are no hotspots, but is prone to deadlocks which require the calling
application to retry failed requests.
LMDB is storage engine optimized for read. Why you don't optimize it's
write using cache oblivious data structure, for example LSM tree or
create new, like in sophia (sphia.org) key value DB?
I just looked, sophia is nothing special. See these microbench results.
http://pastebin.com/cFK1JsCN
LMDB's codebase is still smaller and faster. Nothing else touches LMDB's read
speed.
--
-- Howard Chu
CTO, Symas Corp. http://www.symas.com
Director, Highland Sun http://highlandsun.com/hyc/
Chief Architect, OpenLDAP http://www.openldap.org/project/
_______________________________________________
sqlite-users mailing list
sqlite-users@sqlite.org
http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
