besides DNS, UDP is|maybe needed in the following system: cluster: I am not sure if we can handle >2Gbps inter cluster tcp traffic, maybe we need to think of the UDP implements of the intercluster communication. ICP: it is udp based.
prefetch:
/* This lock fails quite often. This can be expected because,
multiple threads try to append their buffer all the the same
time to the same connection. Other thread holds it for a long
time when it is doing network IO 'n stuff. This is one more
reason why URL messages should be sent by UDP. We will avoid
appending small messages here and those URL message reach the
child much faster */
stream media:
stream is one of our most valuable design, you can not find out any
other system that can handle stream media well and flex. from my
side, we'll take very serious look if we can rebuild the stream
system in y2012.
In all, I can not agree on kill most of the stuff from the core, if it
will break the stream, or we have to rebuild it later. If it does no
harm on ohters, and no one will take care of it for, why not just leave
it there?
在 2011-12-30五的 13:55 -0700，Leif Hedstrom写道：
> Hi,
>
> what do you guys think about eliminating the UDPNet stuff from the core
> and APIs (there are some old, "private" INK APIs related to UDP). I
> suspect most (all?) of this is old code, related to the streaming
> protocols that the core used to support. The only UDP code we have
> (afaik) is DNS related, and I don't think it uses these APIs / classes?
>
> Thoughts?
>
> -- leif
>
--
zym, Zhao Yongming.
aka: yonghao @ taobao.com
signature.asc
Description: This is a digitally signed message part
