Losing data over VPN
Sometimes change is good, sometimes it's bad, and sometimes it's a mixed bag. Such was the case recently when a customer switched a remote office from a 56K line to using a Kerio VPN over Internet connections. The 56K line was slow and expensive, so the switch had immediate benefits. Unfortunately a problem also appeared immediately: an Okidata printer couldn't print more than six or seven pages at a time.
The printer was set up as a shared printer hanging off a PC. As they had a spare HPJetDirect printserver hanging about, my first thought was to throw that in, but nothing changed: after six or seven pages, the printer would still stop.
My next thought was that this might be a MTU problem. I've seen that now and then with VPN's and also with printers, so it was worth a shot. As there were other branch offices not experiencing this problem, I felt the problem had to be at the office with the troublesome printer, so I went there to experiment.
You can set the MTU of a Windows NIC with a registry key: it's down under HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\Interfaces. The MTU key may not be there, but you can add it to the correct card under interfaces. An easier way is to download "DR TCP" from DSLReports Tests and Tools and use that to make your tweaks (thanks to Jeff Wadlow from Kerio for pointing that out). I started experimenting, but nothing changed: the printer would still stop when sent a large enough job.
However, while experimenting, I noticed a curious thing: if I pressed the Select button on the printer to set it off-line, and then pressed it again to set it back on-line, printing would immediately resume with no loss of data. This told me that there was nothing wrong with data flowing through the VPN, and indeed the Kerio logs had already confirmed that: the VPN logs showed no dropped packets whatsoever.
The Okidata printer has a 128K buffer. That's more than six or seven pages, but it might be close to its high water mark: the flow control signal gets sent before the buffer is absolutely full to avoid lost data. So what it now looked like to me was that the printer was sending a flow control "stop" signal but the other end was never seeing a "start" after the buffer was bled down. When I cycled the Select switch, a "start" was being sent, and the printing would resume. Probably the 56K connection was slow enough that the buffer never would fill up, so this was never observed before now.
Now the question is "whose fault is it?". It's most likely the printer, but it could be the print server. As having a spare printer isn't a bad idea, I suggested trying that first. They've ordered one, so we'll know soon.
The other complaint at this office was that telnet connections would disconnect after ten or fifteen minutes of inactivity. My suspicion there was a low timeout setting in their emulation software because TCP timeouts would ordinarily be much longer than that. However, I wasn't able to observe this behavior on site: I waited 45 minutes and there was no problem. I noticed that their app displays the current time in its menus; I asked the people on site to note what it says as they walk up to the machine: that will tell me how long it actually takes to lose the connection and may give a clue as to what is actually happening.
Turns out it was about 40 minutes; I must have noted the time wrong and been just under. The 40 minutes is expected, Kerio says: "WinRoute will discard TCP packets after forty minutes of inactivity. If this is the problem then this problem might be solved by either increasing the TCP timeout or by activating a keep alive feature in the telnet client so that the inactivity timeout is repeatedly reset."
Kerio support provided this:
1. Stop the WinRoute engine
2. Go to C:\Program Files\Kerio\WinRoute Firewall
3. Edit the winroute.cfg file with notepad
4. Search for DefaultTcpTimeout
5. Set DefaultTcpTimeout to a higher value
6. Save the winroute.cfg file
7. Restart the WinRoute engine
Got something to add? Send me email.
(OLDER) <- More Stuff -> (NEWER) (NEWEST)
Printer Friendly Version