Skip to content

HBase Write Performance (2.0.7)

Jump to bottom
xamry edited this page Mar 8, 2013 · 1 revision

Recently we did performance test over current kundera release. This release enable us to fully utilize usage of filter/filterList support provided by hbase. So here are details (given below) :

  • Enviornment: CentOS 6.2

Bulk Load:

Number Of Records(Single Thread) Kundera (in ms) Native Time (in ms) Datanucleus Time (in ms)
1 154 11 124
1000 2175 439 2493
4000 6396 1765 6796
40000 23572 15262 51403
100000 44817 36750 193289
1000000 441918 375377 NA

Concurrent Load:

Number Of Threads (1 record) Kundera (in ms) Native Time (in ms) Datanucleus Time (in ms)
10 46 18 179
100 288 112 463
1000 2501 879 2525
10000 17566 7991 14899
40000 37745 27825 44823
50000 46446 33757 55026
100000 91013 66903 106193

Concurrent Load + Bulk Load:

Number Of Threads (1000 rec/ thread) Kundera (in ms) Native Time (in ms) Datanucleus Time (in ms)
10 14357 4007 13001
100 48640 37300 89744
1000 478872 365430 1044850

Source Code: Source code for producing these results is available here

  1. For kundera and native hbase
  2. For datanucleus hbase.

Steps to run kundera & native hbase jar:

  1. Download jar.
  2. java -jar -Xmx1024m -XX:MaxPermSize=128m (e.g. java -jar /home/impadmin/workspace/kundera-hbase-2.0.7-perf.jar -Xmx1024m -XX:MaxPermSize=128m

Steps to run datanucleus performance jar:

  1. Download datanucleus-performance.zip.
  2. Extract zip.
  3. java -jar {Number of records} {client type} {process mode} { number of threads} -Xmx1024m -XX:MaxPermSize=128m (e.g. java -jar /home/impadmin/workspace/datanucleus-hbase-perf.jar 1 dataNucleus b -Xmx1024m -XX:MaxPermSize=128m

{client type} => datanucleus {process mode} => 1)c(Cnoncurrent) 2) b(Bulk) 3)cb(Concurrent+Bulk)

Disclaimer:

Above mentioned results are on the specified environment and use case and might have different outcome in other cases

HOME

Home

Clone this wiki locally