hadoop-common-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apache Wiki <wikidi...@apache.org>
Subject [Hadoop Wiki] Trivial Update of "Hbase/NewFileFormat/Performance" by stack
Date Sat, 07 Feb 2009 01:26:38 GMT
Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Hadoop Wiki" for change notification.

The following page has been changed by stack:

  09/02/06 22:44:20 INFO hbase.RFilePerformanceEvaluation: Running SequentialReadBenchmark
for 100000 rows.
  09/02/06 22:44:23 INFO hbase.RFilePerformanceEvaluation: Running SequentialReadBenchmark
for 100000 rows took 2840ms.}}}
+ == 16 concurrent reading threads ==
+ For ten byte cells and 8k rfile blocks, against localfs, MapFile wins.  Results for rfile
are odd. Pread runs are slower than seek+read.  Localfs should just be discounted as wonky.
 On hdfs, pread is faster than seek+read for rfile -- about 3 times faster -- and rfile is
about 3/4 times faster than mapfile in above various tests.  My guess is that we pay for the
high-level synchronizations on mapfile when operating with the relatively high-latency hdfs.
+ For 1k cells and 64k rfile blocks, rfile is about twice mapfile on random accesses.  Scan
tests come in closer together with rfile about 20% faster on the seek and read 30 rows test.
 Odd is mapfile beat rfile in the sequential read benchmark -- you'd think rfile will kill
it in this test of all of them.

View raw message