GRUU again for the last time really Jonathan Rosenberg Cisco - - PowerPoint PPT Presentation

gruu
SMART_READER_LITE
LIVE PREVIEW

GRUU again for the last time really Jonathan Rosenberg Cisco - - PowerPoint PPT Presentation

GRUU again for the last time really Jonathan Rosenberg Cisco Changes from -11 Minor Changes Removed references to provisional responses for non-INVITES Added note from Dales draft about not mucking with Contacts with ;gruu


slide-1
SLIDE 1

GRUU

again for the last time really

Jonathan Rosenberg Cisco

slide-2
SLIDE 2

Changes from -11

  • Minor Changes

– Removed references to provisional responses for non-INVITES – Added note from Dales draft about not mucking with Contacts with ;gruu – Added note from Dale’s draft about treating GRUU opaquely and not modifying URI parameters

  • Updated appendix to use EKRs stateless

tokens draft

slide-3
SLIDE 3

Main Change: The Register Race

REG 200 OK Registrar Changes Expiration to 5s NOTIFY reg-info 200 OK REG 200 OK expiration new reg NOTIFY reg-info 200 OK Registrar sends a single consolidated state update for the result of the previous two events since they came close together So did My register beat the clock??

slide-4
SLIDE 4

General Problem

  • IF the registrar is going to muck with

registrations, we need a reliable way for client to synchronize its temp-gruu with server at any time

– Determine whether its set of valid temp-gruu match the servers

slide-5
SLIDE 5

Solution in GRUU-12

  • If server modifies expiration times, MUST

implement reg-event and gruu extension

– Client that implements reg-event and GRUU MUST support gruu-reg-event

  • Reg-event notifications contain CSeq of first

contact registered to that gruu with a given call- id

  • Client keeps all temp-gruu learned in

REGISTER responses whose

– Call-ID matches that of reg-event – CSeq is higher than that of reg-event

slide-6
SLIDE 6

Side Effect

  • Changing Call-ID in REGISTER refresh

will

– Expire all previous temp-gruu – Not cause any outage in registration – just a refresh – low overhead

  • Provides a crude bulk invalidation

technique

slide-7
SLIDE 7

Consequences

  • If an instance fails and recovers and

registers with a new Call-ID and same instance-ID

– All temp-gruu are invalidated

  • Stateless token generation needs to use

Call-ID and original CSeq instead of wallclock time

slide-8
SLIDE 8

Other ML Issues

  • Jeroen: temp GRUUs can be long

– Will note this

  • Ekr-1: Is Supported or +sip.instance the

trigger for GRUU generation?

– As it is currently - +sip.instance.

  • Ekr-2: inclusion of pub-gruu and temp-

gruu in REGISTER MUST NOT or SHOULD NOT?

– Current SHOULD NOT is fine

slide-9
SLIDE 9

Other ML Issues

  • Ekr-3: Should a rebooting UA remove a stale contact?

– No, as it is now

  • Ekr-4: Separators needed in string for cookie?

– Yes – will add

  • Jeroen-1: only do GRUU processing for non-expiring

contacts

– Yes – will add

  • Jeroen-4: What if there are multiple contacts with same

instance and no reg-id (GRUU but not sip-outbound)

– Need to update – use outbound-like behavior to take most recent contact

slide-10
SLIDE 10

Next Steps

  • I have incorporated all comments into a

working -13

  • Once blackout period ends, -13 can be

submitted and go to IESG