[LEAPSECS] stale leap second information

[Rob Seaman]

On Jan 17, 2015, at 4:12 PM, Poul-Henning Kamp <phk at phk.freebsd.dk> wrote:

> If you need the historical information, you can afford the TCP
> connection to IERS.

There's also value in implementing a single ICD for all use cases, but point taken.

> I'm focused on how NTP/PTP clients will be able to get/validate leapsecond information efficiently and reliably.

The python script tests the encoded values:

% ./leap.py
YYYY MM before  after  encoded crc IP               Decoded
-------------------------------------------------------------------------
1972  1      9     10 f0002500  fb    240.0.37.251 -> OK  1972  1   +9 +1
1972  7     10     11 f0182900  b1   240.24.41.177 -> OK  1972  7  +10 +1
1973  1     11     12 f0302d00  45    240.48.45.69 -> OK  1973  1  +11 +1

but doesn't pair them with string addresses.  Were you thinking like Warner suggested, requesting feedback for a specific month/year?  So a client would poll at startup / reset / etc, but would need to understand the June/December convention (and ignore the other technically legal months)?

>> Unless there is an overriding value in efficient encoding, 
> 
> There is.  getaddrinfo(3) is almost always available and works
> about as well as any protocol on the internet can work.

Doesn't getaddrinfo support IPv6, too?  Parsing the longer string won't be dramatically slower.

Rob