Re: Keep alive

Scott and James

If I may be of service on this one. I sincerely apologize if this is inappropriate for me responding here.

The socket options for keep alive do not work in this situation. Client access uses the timing marks from everything I was able to figure out. Never could figure out how though. After a couple of days of sniffing and parameter negotiation tweaking I just gave up and said I would come back to it later. That never happened.

Without mucking up the connection you can use an "Are You There" by sending the following at an interval of no activity:

byte[] b = { (byte) 0xff, (byte) 0xf1 };

This is just a NOP instruction from the rfc 854

NOP 241 No operation.

The rfc specifies 0xF6

Are You There 246 The function AYT.

But the NOOP seemed safer to me and worked.

Hope this helps.

Regards

Kenneth


Scott Klement wrote:

James,

setsockopt() should have an option called SO_KEEPALIVE to enable keepalives at the TCP level. Is this really what Mark is looking for, though? I mean, I know they have the same name -- but are they the same thing?

Seems to me that TCP keepalives are really intended for detection of broken connetions, and aren't sent often enough to keep a telnet alive.

Telnet (RFC 854) also defines an "Are You There" (AYT) code that can be sent periodically to verify that a telnet session is still active -- and there's also an extension to telnet (RFC 860) for timing marks, which has a similar purpose AFIAK.

I'd be nervous about trying to implement any of these without first determining what exactly the other emulators send when they're trying to keep the connection active....


James Rich wrote:

Interesting to note that I usually have the exact reverse problem: Windows Client Access sessions drop but tn5250 marches on perfectly.

In linux there are three tcp keepalive settings: tcp_keepalive_intvl, tcp_keepalive_probes, and tcp_keepalive_time in /proc/sys/net/ipv4 (at least there are on kernel 2.6.24.5). Might these be something like what you are looking for? I thought I remembered seeing an option to setsockopt that would set the tcp keepalive per socket but a quick review of the man page turns up nothing (other than referencing the /proc settings I listed above).

James Rich

if you want to understand why that is, there are many good books on
the design of operating systems. please pass them along to redmond
when you're done reading them :)
- Paul Davis on ardour-dev