• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1February 2002:
2
3The mDNSResponder code has a slight architectural change to improve
4efficiency.
5
6The mDNSResponder code previously called ScheduleNextTask() after every
7operation, to calculate the time at which it needed to be called back to
8perform its next timed operation. When the workload is light, and
9protocol operations are rare and far apart, this makes sense.
10
11However, on networks where there is a lot of mDNS traffic (or the CPU is
12slow), this leads to the following anomolous behaviour: mDNSResponder
13spends a lot of CPU time working out what to do next, when what it needs
14to do next should be obvious: Finish processing the big backlog of
15packets that have been received.
16
17To remedy this, mDNSResponder now only executes ScheduleNextTask() when
18there is no other obvious work waiting to be done. However, the
19mDNSResponder code does not have direct access to this knowledge. Only
20the platform layer below knows whether there are packets waiting to be
21processed. Only the client layer above knows whether it is in the
22process of performing a long sequence of back-to-back mDNS API calls.
23
24This means that the new architecture places an additional responsibility
25on the client layer and/or platform support layer. As long as they have
26immediate work to do, they should call the appropriate mDNSCore routines
27to accomplish that work. With each call, mDNSCore will do only what it
28immediately has to do to satisfy the call. Any optional work will be
29deferred. As soon as there is no more immediate work to do, the calling
30layer MUST call mDNS_Execute(). Failure to call mDNS_Execute() will lead
31to unreliable or incorrect operation.
32
33The value returned from mDNS_Execute() is the next time (in absolute
34platform time units) at which mDNS_Execute() MUST be called again to
35perform its next necessary operation (e.g. transmitting its next
36scheduled query packet, etc.) Note that the time returned is an absolute
37time, not the time *interval* between now and the next required call.
38For OS APIs that work in terms of intervals instead of absolute times,
39mDNSPlatformTimeNow() must be subtracted from the absolute time to get
40the interval between now and the next event.
41
42In a single-threaded application using a blocking select() call as its
43main synchronization point, this means that you should call
44mDNS_Execute() before calling select(), and the timeout value you pass
45to select() MUST NOT be larger than that indicated by the result
46returned from mDNS_Execute(). After the blocking select() call returns,
47you should do whatever work you have to do, and then, if mDNS packets
48were received, or mDNS API calls were made, be sure to call
49mDNS_Execute() again, and if necessary adjust your timeout value
50accordingly, before going back into the select() call.
51
52In an asynchronous or interrupt-driven application, there are three
53places that should call mDNS_Execute():
54
551. After delivering received packets, the platform support layer should
56call mDNS_Execute(), and use the value returned to set the platform
57callback timer to fire at the indicated time.
58
592. After making any mDNS API call or series of calls, the client layer
60should call mDNS_Execute(), and use the value returned to set the
61platform callback timer to fire at the indicated time.
62
633. When the platform callback timer fires, it should call mDNS_Execute()
64(to allow mDNSCore to perform its necessary work) and then the timer
65routine use the result returned to reset itself to fire at the right
66time for the next scheduled event.
67