Ah well keep in mind that going through the Thrift server requires one more network roundtrip, and if you used scanner caching in java then you also need to configure it for thrift. Moreover, you should also do the scannerGetList call and specify more than 1 row to minimize the number of RPCs. Still, it will be slower. What you can also do to help is to use the thrift native PHP extension (google that) which should give you another speedup but still, it will be slower than pure Java since it still requires you to go through Thrift before hitting the region servers.
J-D 2011/6/23 Frank Liu <[email protected]>: > Thanks for your reply. > > Could you provide any advices for locating the problem? > coz I don't know it's because of PHP or hbase thrift server. > > looks when I use single key via php, it's fine. when the records number > increased, it turns slow. > the response time is based on the records returned. > > Frank > > > At 2011-06-24 01:08:23，"Jean-Daniel Cryans" <[email protected]> wrote: > >>30 seconds would point to something in your environment, not a >>configuration issue. >> >>J-D >> >>On Thu, Jun 23, 2011 at 1:35 AM, Frank Liu <[email protected]> wrote: >>> Hi, >>> I have 4 nodes hadoop/hbase environment. >>> When I use Java client to access hbase, works ok. >>> When I use PHP to access hbase via thrift server, the functions is ok, but >>> it's very very slow. >>> I use same startkey and endkey to scan same table in hbase. It takes php >>> about 30s to get >>> the result, but using java client, the result returnedimmediately. >>> >>> Is there any extra configurations needed for thrift server? >>> Thanks! >>> >>> the following is the environment information >>> version: cloudera cdh3u0 >>> node:4 nodes with 4G memory >>> configuration: all default config >>> >>> Frank Liu >>> >>> >>> >
