Anybody here know where I can get help with a TCP communication problem?

We have a terminal emulator problem.

On a handful of hosts, our Java-based TN5250 emulator has a chronic problem of sporadic crashing.

On the host where that happens the most, it happens both with and without SSL. The owners of the host have filed a PMR, and the IBM analyst swears up and down that we are sending 200kB (and larger) packets that are causing buffer overruns.

We have data stream capture and buffer-dump diagnostics built into the emulator, and we have yet to find a single case in which the last thing we sent before the crash (or indeed, anything we sent) was even as much as 16kB, at least as an RFC 1205/2877 TN5250 data stream passed to the socket.

Moreover, according to the traces and dumps on our end, there are at least as many cases in which the last traffic through the socket was sent FROM the host TO the emulator.

Almost from the beginning, it has looked to me like some sort of network issue. Except that IBM's own emulator is apparently immune.

It's driving me crazy. Every time I see new email from the customer with the open PMR, my blood starts to boil. And none of the Google searches I've tried, looking for some sort of list-server or board for help diagnosing Java SSL problems have turned up anything even vaguely relevant (most seem to be concerned either with certs, or with HTTPS, or both.

When we establish secure connection, we start with a static SSLSocketFactory. This is instantiated using an SSLContext that has a very basic TrustManager. We then, whenever a new secured socket is needed, get the socket from the factory with a "createSocket(adr, port)" method call. Perfectly straightforward, and it works just fine on the overwhelming majority of hosts.

I don't understand what could make the Telnet server think it was getting a 200kB packet, when I can find no evidence that we've ever sent anything remotely that big.

--
JHHL