[EAS] CAP Weekly Test on the 8th

[Sean Donelan]

On Tue, 16 Oct 2018, Adrienne Abbott wrote:
> So far, this process is working well for our stations. Of course, this delay
> does not apply to the EAN and NPT, only local/state CAP EAS activations.

Adrienne is correct. Coordination among participants in an area is 
needed. Since FEMA hasn't published any performance parameters, Bush's
EO 13407 revoked the Clinton White House Statement of Requrements, 
everyone is just guessing what's needed. Given that, here's some 
suggestions.

Content Distribution Networks used for IPAWS have built in web caching 
intervals. The first poll of a CDN cache will often result in a cache 
miss, and they have DDOS defenses, so extremely short polling intervals 
are self-defeating. You want to coordinate Local/State Primary EAS station 
delays with IPAWS polling intervals.

Suggested CAP polling intervals and LP/SP delays:

I rounded the numbers for simplicity.

CAP polling interval = 60 seconds, LP/SP station delay = 200 seconds
CAP polling interval = 30 seconds, LP/SP station delay = 100 seconds
CAP polling interval = 10 seconds, LP/SP station delay = 50 seconds

Participating National cable and broadcast stations without downstreams 
don't need to add any delays. They should use the same CAP polling 
interval as their upstream sources.

Due to CDN caching, CAP polling intervals less than 10 seconds don't 
seem to improve things due to cache misses.

SP/LP delay is measured from reception of first EAS header preamble or CAP 
XML message (not including audio file download) to start of LP/SP on-air 
transmission of preamble EAS header. If your LP/SP transmission system 
has encoding or switching delays, you need to account for those.

In my testing and queue modeling, this prefers the CAP message over 99% of 
the time. In case of Internet outages, backup over-the-air EAS distribution
still has reasonably short delays, assuming states only have one or two 
hop EAS daisy chain. States with long multi-hop EAS daisy chains should 
shorten those chains.

Problems:

PEP stations also need to be delayed after IPAWS releases the CAP message.

Some EAS boxes can only buffer a single EAS message, which can be 
overwritten by too long delays.

NOAA assumes SAME messages can be transmitted back-to-back, with no 
intervening delay. 60 second CAP polling needs minimum of 200 second 
inter-message over-the-air EAS/SAME inter-message gap. If NOAA and FEMA 
ever work out the issues with distributing weather warnings through IPAWS, 
this will become an issue.

EAN message buffers may be less than 120 seconds, so any transmission 
delays must be shorter than the audio buffer size.

Improvements:

Triggered or accelerated polling when an EAS header is received reduces 
some delays. However, a loosely coupled network system is more reliable 
than a tightly coupled network system.

Suggested software changes requiring FCC approval when an EAN message is 
received:

Assumptions, until FEMA publishes actual performance parameters.

When an EAN header received, switch to accelerated IPAWS polling, but do 
not immediately poll to avoid network self-synchronization and congestion.

EAN accelerated CAP polling parameters:

All times measured from successful decoding of EAN header, not including 
EAS attention signal.

a. audio buffer continues to capture from EAS source, and does not lose 
any part of the audio while polling IPAWS.
b. 10 second polls with randomized jitter to avoid network synchonization
c. up to 4 polling attempts
d. maximum download delay of 45 seconds from decoding of EAN header, 
including DNS, TCP, TLS delays
e. assume 2 minute audio file retrieved until EAN streaming defined

If CAP message ?or/and? audio file cannot be retrieved within 45 seconds, 
use the audio buffered from the EAS monitor source.

Assumption: Goal for presidential audio message "on-air" in less than 60 
seconds of initial EAN notification, including manual board operator 
action, EAS headers/attention tones, audio/video encoding, switching, 
transmission and other system delays.