Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
HBase, mail # dev - Reads are very slow


Copy link to this message
-
Reads are very slow
Vibhav Mundra 2013-05-21, 10:21
Hi All,

 I am trying to use Hbase as a key-value store, where the key is stored as
row-key and the value as a column family.

The desc of the table is as follows:
 {NAME => 'cookies', FAMILIES => [{NAME => 'cf', DATA_BLOCK_ENCODING =>
'NONE', BLOOMFILTER => 'ROWCOL', true

  REPLICATION_SCOPE => '0', VERSIONS => '1', COMPRESSION => 'GZ',
MIN_VERSIONS => '0', TTL => '214748364

 7', KEEP_DELETED_CELLS => 'false', BLOCKSIZE => '8192', IN_MEMORY =>
'false', ENCODE_ON_DISK => 'true',

  BLOCKCACHE => 'true'}]}

I am accessing the values using rest API.

curl -H "Accept:application/octet-stream"
http://10.4.41.133:10000/cookies/ed36b1d218315_ziv/cf:data

This is bound to return a single result, since it used as key-value pair.

Now the strange part is sometimes my request takes 40 millisecs to answer.
I am firing some 50000 request per second. Is there something I can do to
optimize. Basically I am looking at returning the results in 5 millisecs.

The hbase configuration is as:
1 node--> running hbase master and hadoop-namenode/secondary node and hbase
rest client.
+
2 nodes--> running data-nodes and regionservers.

The 2 nodes that are running region-servers/datanodes are having 4 ssd's
each and the data is spread across all the four ssd's.
Each node has a 12 GB RAM.

Also I find this very strange, there is negligible disk I/O, CPU usages on
datanodes.

I have tried various combinations, but none of them have given any results.
I am attaching the hbase-site.xml. If required I will also attach the
hadoop configuration files also.

-Vibhav