[PATCH] Workarounds for lotro bugs

[Daniel Santos]

Thanks for the feedback.  I haven't gotten to that level of analysis 
yet, because I was more focused on a nice working solution up front, but 
will start digging deeper.  Really however, the call to Sleep() is just 
an added bonus.  The fix works fine without it, you just get some frame 
rate drop whenever the client is acting up (I have a quad core, so I 
don't know what the impact is for dual or single core users).  I would 
be happy to leave it out if we could get a fix in the wine mainline for 
now while digging deeper into what's causing it.

The main problem is the production of thousands of small and identical 
udp packets essentially causing a DoS attack on their server and bogging 
down the upstream traffic for the user.  In one test, 8500x 62 byte 
packets (size includes headers) were sent in 500 mS.  Some users have 
also reported problems where their routers don't seem to handle the 
flood very well (crashing, etc.).  It's an intermittent problem and only 
occurs while one's character is in specific locations in the game 
world.  I'm also not certain that every Linux user experiences this 
problem and the degree will certainly vary depending upon their bandwidth.

Rob Shearman wrote:
> 2008/7/16 Daniel Santos <daniel at bytesage.com>:
>   
>> +    if(_lotro_hack_enabled
>> +            && to->sa_family == AF_INET
>> +            && dwBufferCount == 1
>> +            && _lotro_hack(&get_per_thread_data()->lotro_msgs,
>> +                    s, lpBuffers, (struct WS_sockaddr_in*)to)) {
>> +        Sleep(10);
>> +        if(lpNumberOfBytesSent) {
>> +            *lpNumberOfBytesSent = lpBuffers->len;
>> +        }
>> +        return 0;
>> +    }
>> +
>>     
>
> Hi Daniel,
>
> While I appreciate the effort you've gone to to fix a bug, I don't see
> any analysis in the bug that suggests this is the only possible
> solution. It seems to me as though LOTRO is calling WSASendTo in a
> tight loop and that it isn't such a tight loop on Windows. A +relay
> log will reveal which calls it is making in that thread. You can then
> build a similar test program and time how long each call takes on
> Windows and Wine and see where the difference is (although it probably
> won't be as simple as that as there will be other threads interacting
> in LOTRO). At a guess, it might simply be that Windows yields to other
> threads after making a call to a network function.
>
>   

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.winehq.org/pipermail/wine-devel/attachments/20080716/7ccb5e67/attachment.htm