<p dir="ltr">I wouldn't mind if it were made configurable... </p>
<p dir="ltr">And I note that ssh in particular sets the old style TOS imm bit so I doubt that anybody fails to pass traffic with that set.</p>
<p dir="ltr">So perhaps your failing corporate tester could try patching mosh to use that instead. Or fix their broken network.</p>
<p dir="ltr">I note moving to best effort or to the imm bit will move mosh traffic out of the Linux WiFi VI low latency queue and into the bloated best effort queue and your users might notice a large loss in performance... But that would be good to evaluate... And thank you for the research into the effects of classification on mosh traffic regardless.</p>
<div class="gmail_quote">On Mar 10, 2013 4:41 PM, "Keith Winstein" <<a href="mailto:keithw@mit.edu">keithw@mit.edu</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Hi Dave,<br>
<br>
Sorry to report that we got another complaint that AF42 datagrams were<br>
being dropped by somebody's corporate network. I think I am going to<br>
roll back to just ECT for the next Mosh release and see if these go<br>
away.<br>
<br>
Ideally we would have a more sophisticated scheme where we could try<br>
AF42 but then fall back to no DiffServ codepoint after a while if that<br>
failed, but we aren't quite there yet.<br>
<br>
Hope all is well,<br>
Keith<br>
</blockquote></div>