DAOC TCP has bugs, causes needless link deaths.Follow

I've tried support@DAOC, the feedback forms on herald, and am running out of ideas as to where to send this information to get a response.

Essentially while trying to diagnose whether my firewall or my ISP were dropping data causing my link deaths, I put a lan analyzer on either side of my firewall while running DAOC. Indeed I get occasional dropped packets coming from my ISP(once every few minute or so). What is disturbing is that a single dropped packet on the TCP socket DAOC uses for part of its communication is sufficient to cause a link death. The DAOC client appears to recognize the missing packet from the server, and sets its TCP ack to point to the missing packet, thereby requesting the server retransmit it. The server responds to each such request with an acknowledgement making it clear that the request was received, but does NOT resend the requested packet. This is bizarre because this retransmit behavior should be part of the operating system on the server, and not even part of the DAOC server application itself. After 30 seconds of requesting a retransmit, the client eventually closes the socket and exits to the link death screen.

I've repeated the experiment with the lan analyzer running 5 times now, and every time is the same. Despite the fact that client and server are still exchanging messages, the client has to exit 30 seconds after a packet from the server is dropped because the server fails to restransmit the missing packet.

This is absurd. This should not happen. Mythic should be able to fix, but I can not figure where to go with the information to get something done about it. Any other internet application can handle missed packets like this so my ISP has no motivation to fix the dropped packets. But even if they fixed it I would not expect any internet link to be completely without dropped packets.

This is _so_ frustrating. I have multiple sets of lan logs I could give to developers to help diagnose problem, and can't even get a response.

I know my ISP has 400 users on one cable. Probably with one router at the head of the cable. I have every reason to believe that my LD problems are frequent due to that.

This doesn't change the fact that a single packet, packet loss should not cause an application using TCP to fail, because TCP has built in recovery.

I've been working this problem for a month and have indeed looked at the tech support articles on the herald support page.

1) Its not the firewall. Still have the problem if bypass the firewall entirely. Can see that the packet was lost from the server size of the firewall(between firewall and modem), indicating it was before it got to firewall.

2) Tried the 10/100 half/full duplex change and makes no difference. Lan logs indicating the loss is detectable on other side of firewall on inbound packets would seem to confirm that the connection from PC to firewall is not the issue.

The problem very much seems to be the server is not retransmitting a lost packet when the client asks for it. Despite every evidence that the server did get the request for the lost packet.

Only parameters I cand find that would affect this would be server side parameters. Some O/S's allow tweak of the retransmit timeout and of the number of requests required prior to retransmitting a packet prior to the timeout. Both of these parameters would have to have been set absurdly large for the server to not retransmit the packet after 30 seoncds of requests at around 2 requests per second. The retransmit timeout should be computed automatically from the estimated round-trip time. I suppose there is could be some oddity causing the round-trip time to be miscomputed so that the server thinks it needs to wait absurdly long time, but that alone wouldn't change the fact that server should retransmit lost packets if more than three requests for the same lost packet are received.

I can't find anything that I could change on my side that would cause the server to be more likely to retransmit the dropped packet.

I have been playing DAoC for over 2 years, and I have gone LD maybe a total of 10 times. If your going LD that often it's a problem with your ISP or settings.

I'm sure everyone gets dropped packets now and then without going LD - that's part of the nature of the internet, however, as your ISP has "finally fixed something" my guess is that it was doing more than just dropping an errant packet here and there. Maybe it only reported one dropped packet when it had actually dropped a bunch of them? Or, it wasn't accepting retrieved packets. Whatever the problem, sounds like it was the ISP's fault.

Glad to hear you got a happy resolution! Hope you enjoy the game.

ouple pieces of additional information in case someone is actually trying to solve the problem:

1)
My ISP admits they have a minor packet loss problem. Out of 400 users on this cable line, there are 5 complaints. 2 of those are people trying to maintain VPN tunnels, and 3 of them are people trying to play DAOC. Hmm, either speaks to popularity of DAOC, or to it's fragility to packet loss.


2)
In analyzing lan logs I have indeed seen one case where the TCP retransmit recovery functioned correctly and a dropped packet from the server was retransmitted by the server, and the game proceeded for a consderable length of time(several minutes) untill another dropped packet failed to be retransmitted, and resulted in link death after the typical 30 seconds of the client requesting the retransmit.

That's amazing! I'm glad your persistence has paid off. I'll try to remember that if I ever run into someone else with the same problem. I hope your ISP sorts out the problem. If not, I guess you'd better switch to a different ISP :)