Clint Morgan wrote:
Nice work!
Is the random read with BlockCache enabled?
BlockCache is enabled by default on catalog tables only. In the test,
it was not enabled for the user-space PE test table.
My experience with blockcache and random read was that with 1G heap,
tests usually ran slower when blockcache was enabled. Because the test
is 'random', test was spending its time clearing and filling the cache.
If heap was bigger, 2G say, then blockcache made a difference and random
reads ran faster (2-4X faster).
Real loads are probably not random.
If you are considering enabling blockcache on tables, be sure to
consider findings in HBASE-1127; i.e. difficulty making the soft
reference cache run reliably.
St.Ack
On Fri, Jan 16, 2009 at 9:20 PM, stack <[email protected]> wrote:
Here you go J-D:
http://wiki.apache.org/hadoop/Hbase/PerformanceEvaluation#0_19_0.
4-5X improvement writing and scanning. Random reads are up ~25%.
I'm going to start a new table with numbers for concurrent clients so we
can track how this changes over time.
St.Ack
Jean-Daniel Cryans wrote:
I think we're also due for a new PE comparison.
Thx in advance Stack ;)
J-D
On Fri, Jan 16, 2009 at 4:49 PM, stack <[email protected]> wrote:
The first hbase 0.19.0 release candidate is available for download:
http://people.apache.org/~stack/hbase-0.19.0-candidate-1/<http://people.apache.org/%7Estack/hbase-0.19.0-candidate-1/>
<http://people.apache.org/%7Estack/hbase-0.19.0-candidate-1/>
More than 180 issues have been addressed in 0.19.0. The release notes
are
available here: http://tinyurl.com/8xmyx9.
Should we release this candidate as hbase 0.19.0? Please vote +1/-1 by
Obamaday (Tuesday), January 20th, 2009.
Thanks,
St.Ack
