Performance Level

Sustained performance capable of "initial load" inserts to the B-Tree index (sorted on given key (structure)) store of 100 million random records in 35 minutes using a Mac Mini.
i.e. - 47,619 IO per second (IOPS) using commodity personal PC/SSD !

Post load I/Os are at sustained high speed levels. Reading all of 100 million records randomly in minutes. Retrieve in ascending/descending level at the same speed levels.

Parallel Reads are supported with full hardware parallelism where hardware supports it.

Last edited Oct 17, 2015 at 10:47 PM by grecinto, version 3

Comments

grecinto Mar 22, 2013 at 12:07 AM 
Hi Alek,

I'll compare when time permits... Also, sop 4.7 (vNext) will have Mvcc feature for enterprise scale. The current version focus is to be a solid bldg block of an engine.

Thank you,
-G.

stalek Feb 17, 2013 at 4:56 PM 
Hi,
could you compare the performance of your engine with the benchmark below?
http://leveldb.googlecode.com/svn/trunk/doc/benchmark.html
I mean if you could run LevelDB tests from this benchmark on the same machine you did tests of your engine (they provide test source code)? How many operations (inserts) can you achieve per second?

Could you write if you use locking schema like MVCC or something that requires real synchronization on data access level?

Thanks,
Alek