home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
The World of Computer Software
/
World_Of_Computer_Software-02-385-Vol-1of3.iso
/
x
/
xntp3.zip
/
xntpd
/
ASSUMPTIONS
next >
Wrap
Text File
|
1991-09-28
|
2KB
|
49 lines
* (peer->flags & PEER_CONFIG) iff (peer->hmode in {1, 3, 5})
* sys.clock in 3.4.9 == sys.poll in 5.2
sys.update in 3.4.9 == sys.clock in 3.2.2
* It's valid to drop packets from a broadcaster when in active or
client modes
* The new algorithm for dropping packets + the reach/valid rules
subsumes the peer->trust/poll stuff, so it goes away.
* The test for stratum 1 near the end of leap_process is wrong.
* leap indicator, stratum and refid follow sys_peer (best selected
clock). Currently only packets from sys_peer update reftime and the
local clock --- should it be every packet from a selected peer?
* It is no longer a good idea to drop skew update on short intervals
(note that the update is now multiplied with interval length, not
divided as in version 2). This is especially true when there are
multiple update sources (see previous).
* peer.dispersion_{index[i]} means \epsilon_i at bottom of page 35.
* In clock_select, the test on reach subsumes the old tests on trust,
delay and stratum.
* NTP_SKEWFACTOR will stay equal to 16, so the result in a u_fp is
just the second part of the interval.
* The code in appendix I does not add peer->rootdelay at the places
that correspond to calls of dist() in the body of the report; it
doesn't take the absolute value either. We do both, anyway.
* Also, even though appendix I calls combine(), it calculates system
variables based on sys_peer values (including offset), so that's
what we'll do too.
Though we'll use sys_peer's select dispersion instead of the
largest retained one.
* The test for "peer passes test 1 to 8" is implemented as a test on
peer->reach, because no indication is kept of a data invalid event.
By the way, what happens if a peer in the synch. set starts failing
data validity consistently? No calls to clock_filter, so it'll just
stick there until the skew forces its dispersion high.
* The descriptions for peer selection status 2 and 3 are weird. For now,
2 means goes into intersection, 3 means survives.