I had the same issue. I ended up writing a coprocessor that did this under
the BatchMutate hood using a special attribute in the Put operation, and
that increased performance significantly to be almost equal to the regular
batchMutate(Put) version.
On Sat, Jul 9, 2016 at 7:59 AM, 陆巍 wrote:
> The
Note that you're doing single RPCs for each Increment. You should batch them
(see HTable.batch(...))
What version of HBase? What's the ping time between the client and server
machines?Is Nagle's disables (tcpnodelay set to true)? Otherwise each RPC will
take at least 45 ms.
Lastly you'll likel
Which release of hbase are you using ?
Does it include HBASE-15213 ?
Thanks
On Sat, Jul 9, 2016 at 3:14 AM, 陆巍 wrote:
> Hi,
>
> I had a test for Increment operation, and find the performance is really
> bad: 94809ms for 1000 increment operaions.
> The testing cluster is pretty small with only