Re: [reSIProcate] Timers: why system time?
Byron Campen wrote:
Let's say one of these timers is at the top of the heap, and
whatever error you're thinking about happens. Are you saying that not
only will that timer be 49 days late, but everything else in the heap
will be too?
Not new timers. But, if we miss the rollover, then everything in the
heap at that time will instantaneously appear to be 49.7 days later than
their original expiration time. New timers will be inserted in the heap
just fine (ahead of the "lost" timers), and expire correctly.
The behavior would be similar to using our current stack and setting the
system time to be 49 days in the past while timers are pending.
/a
Best regards,
Byron Campen
Technically, they won't be "lost" as much as "over 49 days late."
From a SIP protocol perspective, 49 days is long enough to consider
the timer lost for all practical purposes.
/a
Byron Campen wrote:
I can see how we would miss a roll-over with low timer activity,
but I am not sure what you mean by "lose timers". If we have lost a
timer from a timer heap, that implies the heap is now broken,
because that is the only way to "lose" anything. So this is either
much more serious, or much less serious than you have stated.
Best regards,
Byron Campen
Scott (and any other Windows developers on the list):
The Windows 49.7-day rollover fix in this patch uses heuristics to
handle the rollover in an efficient manner -- you can actually lose
timers if you have relatively low timer activity and a multi-minute
timer running when the tick count rolls over. The chances of this
happening are, admittedly, vanishingly small (especially in normal
SIP usage). In any case, you might want to give that code a quick
review to ensure that you're comfortable with it.
Also, I think it's safe to say that it's difficult to test the
correctness of the code, as step one of any such test plan
necessarily involves something like: "boot a Windows machine and
wait 49.6 days." Make sure you're comfortable with that fact as well.
/a
Alexander wrote:
Hi
Here is a patch to use monotonic clock.
It is 3 modified files from reSIProcate 1.4.
Attached files have Windows (CLRF) eof-style.
Windows
=======
GetTickCount() with 16 ms inaccuracy used.
It is very simple to change it to timeGetTime() etc with 1 ms
accuracy,
but it will require init/fini calls on application level and linking
with Winmm.lib
Tested on Windows XP SP2, compiled with VS7.1
Not tested on Windows CE
Linux
=====
If monotonic clock is not available (compile time and runtime) it
will
fall back to gettimeofday()
I am not familiar with reSIProcate build system so I use
workaround to
avoid linking with librt:
syscall( __NR_clock_gettime, ... )
instead of clock_gettime(...)
IMHO it may not work on other POSIX compliant platforms.
Pass only some basic tests on my Linux PC (2.6.18 kernel)
Any feedbacks appreciated.
Regards
Alexander Altshuler
Xeepe team
http://xeepe.com
------------------------------------------------------------------------
_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel
_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel