Weird, leveldb typically has better perf than kahadb. Then again I typically use the JNI leveldb driver when I benchmark it so perhaps thats the big difference. But your benchmarking case also is a bit different from what I've looked at. Any chance you could post a little github project /w the scripts needed to reproduce your benchmark?
I'd like to give it a try on my hardware. On Wed, Nov 6, 2013 at 3:37 AM, Attila Nagy <[email protected]> wrote: > Hi, > > AMQ 5.9.0, FreeBSD 9/amd64, openjdk version "1.7.0_25". > Publishing to a virtual topic, with one durable subscriber disconnected, 1 > million small (160-300 bytes payload) messages, with some headers, then > consuming with one client. > The queue has a 10MB memory limit, store cursor. > > With kahadb (default settings) it takes 239 seconds to publish, 368s to > consume (4184, 2717 messages per sec average). > With leveldb (default settings, native) it takes 451 and 488 secs (2217 and > 2049 msg/s). > > What's recommended here? > > Thanks, -- Hiram Chirino Engineering | Red Hat, Inc. [email protected] | fusesource.com | redhat.com skype: hiramchirino | twitter: @hiramchirino blog: Hiram Chirino's Bit Mojo
