Problem in writing to Mainframe Server through a Socket

Question

We have a Java client application running on MS windows writing/reading from a Mainframe application.
What we observe is that, on some client machines even if we do not explicitly convert to EBCDIC the data gets sent in proper EBCDIC format, but on some machines the data that is sent is not in EBCDIC format. We have the same code on both the machines. Also one more thing that is coimplicating is that on the machine where explicit conversion to EBCDIC is needed now, used to work fine without this conversion about 3 months back. The token we write to socket is TSSK
Is there some change in Java releases that has caused this more stringent checking now? Or are there any changes in Microsoft releases that is now more particular about the conversion?
Your input is greatly appreciated. Agree that this might not be a socket issue, but was not clear where to post the problem.

Answer 1

We use Socket as follows:
byte[] b_tssk = new byte[4];
_socket = new Socket(host, port);
in =  new DataInputStream(socket.getInputStream());
out = new DataOutputStream(socket.getOutputStream());
out.write(btssk,0,4);
_in.read(tranid, 0, 10);
Before writing to the socket we convert the byte array to the required format.
When you say protocol does it mean TCP/IP in this case? And how can we configure it and where?
Can you please elaborate more on this?