BIFF Architecture. IETF 69. Lisa Dusseault. Analogy to CPP

1
BIFF Architecture

• IETF 69
• Lisa Dusseault

2
Analogy to CPP
Events
Lisa is busy
Change state statusbusy
IMclient
Lisa is busy
Lisa is busy

• CPP Common Profile for Presence
• XMPP and SIP/SIMPLE so far
• The Email Server is like a Presence-providing
  client
• Email store knows state sends state and state
  changes
• Clients subscribe to specified state
• Start a subscription knowing current state
• Notified when it changes

3
Architecture
CPP System
Clients
Administrative domain
BLAST Protocol
Notification Service
Email Store
XMPP Server
State change UnseenMsgs567
SIP Server

• BLAST sends events once each from mail store to
  notification service
• CPP system can cross administrative domains but
  BLAST does not
• CPP system deals with deliver and storage of
  state for reliable deliver
• CPP system performs the authorization decisions
• Hour glass analogy .

4
Careful about state

• Email server has vast amounts of state
• Delivers relatively small state changes to
  notification server
• Notification server has state
• Latest changed state from email server
• (as well as state of subscriptions)
• Thus, what is a state change for the email server
  becomes state in the presence system

5
Filter Design
Client variation What state information is
interesting How often can receive notifications
Requires Filters

• What is the simplest filter that handles 80 of
  use cases?
• Naming a resource and an event type.
• A resource means a named mailbox
• Type of event could be numberMessageChange,
  quotaStatusChange, etc.
• Thus no SIEVE in SUBSCRIBE in this proposal

6
Reliability Goals/limitations

• Client information cannot be completely bogus
• Use case display 98 unread messages on status
  bar
• Client must be able to detect activity too
• Use case ping when new message arrives
• Limitation Servers will not hold undelivered
  state changes
• They will, however, hold current summarizing
  state
• Schema design the information that must be
  reliable goes into the state, not in the state
  transitions

7
CPP System Reliability

• Client trades off reliability and implementation
  cost
• High reliability is not precluded
• Different protocols and extensions available

8
Event Schema Design

• Very important to achieving use cases
• Need to define state and state changes
• A new message arrived
• -- vs --
• 108 unread messages exist
• -- vs --
• Message waiting indicator should be on