Telnet to 2626 mgmt port freezes

[Baylink]

I'm standing up a stack of 4 used, factory reset, J4900B's.

Web management seems to be working ok, but when I telnet to them, they eventually lose the ability to hear the characters I type -- though the telnet sessions isn't completely dead: if I'm in the menu, the clock keeps counting up.

Each got a fresh IP from DHCP on boot, and has been since been given a known-clean static IP (the network they will service behind the (also new) router, has nothing else on it.  I've put the first switch in Commander Auto-Grab, and then added each of the other three to the stack one at a time, by putting it in Candidate Auto-Join.  I then turned Auto-Grab back off, and my CLI concurs that I have all 4 switches in the stack in the proper order, and the CLI asks me which one I want to talk to when I telnet in. 

But no matter which one I pick, and whether I stay in the CLI or enable the menu, telnet stops listening to my typing after a random period between 5 and 60 second, and never starts again.  The web interface is still functional at this time, and at least once, the telnet session was on the Stack screen, and it showed the first join, but not later ones -- *even though the clock display kept counting*.

The units are running (commander) h.08.83, h.08.106, h.08.83, and h.10.83, all with an h.08.02 ROM.

They've been powered up for over 24h, and other than this, seem to be working and passing traffic fine.

I've seen other reports of this sort of problem in my searching, but none with any useful responses.

I have h.10.117 downloaded; is it worth my time to walk the stack and upgrade them all?

Is there a known solution to this problem?  I've worked with a couple dozen 2626's in my career, and never seen anything remotely similar to this problem.  I'm using Putty 0.65 on Win 7 to telnet to them; they didn't answer the phone at all on ssh (connection refused).  I haven't yet modified the time protocol settings, or anything else like VLAN, just the IPs and stacking.

[Baylink]

I suppose I should say that I'm telnetting to this through a port on one of the switches in the stack, and that the switch is otherwise passing everything I throw through it.

[Baylink]

Further: When the telnet session freezes, I have just learned, a Telnet Break unfreezes it.

It happens too frequently for this to be a serviceable workaround, but perhaps it's a hint.

[Vince-Whirlwind]

How about if you console onto the switches via the console port, do you experience a similar problem?

Give each switch its own IP address and then delete all "stack" "commander" "member" config. Does the problem persist on every switch?