Hi
Thanks for the pointer ashish, 3x performance increase seams like a
bliss :)
Regarding another mater...
Can the client flush on-demand, emptying the client buffer?
I tried bufferedwriterfilter adding it to the last of the filter
chain but for some reason the fw simply stopped sending messages.
Anyhow it would be great if I could flush based on the business logic.
Is this possible?
Regards
On Feb 10, 2009, at 8:40 AM, Ashish wrote:
On Tue, Feb 10, 2009 at 12:07 PM, David Alves <[email protected]>
wrote:
Hi All
I'm developing a massive cluster application that uses Hadoop
RPC
mechanism to communicate.
Due to throughput issues I'd like to change parts of the
communication to async mode and I thought of MINA.
All message objects implement the Writable interface meaning
they can
write/read themselves from a DataOutput/DataInput.
What do you think is the best strategy for implementing the
ProtocolCodecFactory/Encoder/Decoder that would have the least array
duplication/creation?
I thought one of the best options would be to build an adapter
between IoBuffer and DataInput/DataOuput since the methods are
really quite
the same.
Refer org.apache.mina.handler.stream package. May be of help.
What do you suggest? is this the best approach? if so should I
duplicate a class like SimpleBufferAllocator?
Regards
David Alves
--
thanks
ashish
Blog: http://www.ashishpaliwal.com/blog
My Photo Galleries: http://www.pbase.com/ashishpaliwal
