Non-INVITE

1
Non-INVITE

• Robert Sparks
• dynamicsoft
• SIP IETF58

2
Problem Review

• Non-INVITE transactions (NITs) suffer from a race
  condition at the end
• The race renders 408 Useless
• NIT timeouts across multiple proxies result in a
  response storm
• NIT timeouts doom forking proxies
• Delayed NIT responses can blacklist an element

3
The Race Condition
4
408 is Not Useful
UAC
UAS
Request
64T1
64T1
Timeout
408 Request Timeout
5
The 408 Cascade Storm
UAC
Proxy
Proxy
Proxy
UAS
Request
Request
Request
Request
64T1
64T1
408
408
408
408
408
408
408
408
408
408
6
Making it better

• Alternative A Improve the situation while
  keeping the fixed NIT duration
• Alternative B Allow the NIT to pend

7
Alternative B Allow NITs to Pend

• Remove Timer F
• All the above problems are removed
• CANCEL becomes meaningful
• Retransmission roles dont change, so it can be
  chatty
• Backwards compatible
• Forces intermediaries to keep much more state
  than they currently do.

8
Alternative A

• Reduce the probability of losing the race
• Reduce useless traffic
• Address both delayed response and no response
  scenarios

9
Proposal 1 Make the Best Use of Provisional
Responses

• Disallow non-100 responses to NITs
• Allow 100 after it will not damage recover from
  lost messages.

10
Proposal 2 Remove the late response storm

• Disallow 408 to non-INVITE
• Absorb late non-INVITE responses at proxies

11
Proposal 3 Improve UAS knowledge of remaining
time
UAC P1 P2
P3 UAS NI-Timeleft
32 ---
---. NI-Timeleft 29.5
--gt---.
NI-Timeleft 25.5
--gt---. NI-Timeleft 24

--gt---.
--gt


32s

.-200-

.-200-lt--'
.-200-lt--'
.-200-lt--'
lt--'
v
(timeout) ---

12
Proposal 4 Strengthen 3263 caching language

• Make the success caching suggestion normative
  (SHOULD)
• Add caching based on transport failure (including
  non-responsiveness)
• Recommend cache expiration times for each of the
  above

13
Proposal 5 Build a Try again later behavior

• Strawman Proposal
• Redirect to self with a Retry-After indicating
  when the real result might be ready
• Has many issues discussed in draft

14
Recommendation

• Pursue Alternative A, Proposals 1-4 aggressively
  (handoff to IESG before Seoul)
• Research A-5 on a less aggressive timeline
• Abandon B (Should we capture the thinking to date
  and a reason to not proceed in an Informational
  document)?