NTP and PTP Random Thoughts on Trade-offs and Scalability

1
NTP and PTPRandom Thoughts on Trade-offs and
Scalability

• IETF 68
• TICTOC BOF
• Karen ODonoghue
• 1-540-653-1567

2
Objective

• As requested by the BOF organizers
• Kickoff a discussion on the tradeoffs between NTP
  and PTP (IEEE 1588), in particular as related to
  scalability
• Thought provoking slides?
• As revised by the speaker
• Identify some basic metrics for scalability
• Propose a possible way ahead
• Rationale for revision
• May be a bit premature to discuss specific
  NTP/PTP tradeoffs (suffice it to say they exist)

3
Characteristics of Scalability

• What are scalability requirements?
• Number of clients/slaves supported
• Number of servers/masters available
• Number of network or infrastructure devices
• Clock (tod and/or freq) aware (e.g. transparent
  clocks)
• Non-clock (tod and/or freq) aware
• Mixture
• Number and types of networks traversed
• Logical and Geographic distance
• Available bandwidth
• If a potential WG is limited to infrastructure
  networks, how can the requirements be
  narrowed/clarified?

4
Impacts to Scalability

• What impacts scalability?
• Client/server approach versus Master/slave
  approach
• Server/master complexity
• Client/slave complexity
• Hardware requirement
• Network design
• Communications model
• Unicast versus Multicast
• Protocol Operations
• Modes of operation (NTP)
• Types of devices (PTP)
• Algorithm complexity
• Edge cases
• Master/server selection and redundancy

5
Clock Standardization ActivitiesAt a Glance

• IEEE 802.1AS
• Pros
• Hardware time stamp capability available in every
  Ethernet NIC (mass market availability)
• Cons
• Immature at this point
• Targets consumer market (may not be suitable for
  enterprise or telecom infrastructures)

• IEEE 1588 (PTP)
• Pros
• Designed for high precision applications
• Industrial automation and telecom industries have
  real time requirements
• Cons
• Lacks the robustness and deployment experience of
  NTP
• Currently represents a niche market relative to
  generic computing equipment and enterprise
  deployments

• IETF (NTP)
• Pros
• Scalable and robust
• Subnet technology independent
• Represents the generic computing equipment market
  
• Wide scale deployment and years of experience
• Cons
• Does not address high end synchronization
  requirements in an open standard manner

Note Each of these solutions has some overlap
and addresses some unique aspects of the problem
(not an apples to apples comparison).
6

• Given that
• There are various solutions addressing parts (but
  not all) of the problem space
• There is a significant investment in this problem
  space represented by NTP and PTP
• There is an expectation that both NTP and PTP
  will be encountered in the future
• Perhaps
• There are aspects of both that can be identified
  and leveraged for a next generation solution
• At the very least, an interface between the two
  should be considered

7
Possible TICTOC Way Ahead

• Clarify requirements
• Both time and frequency
• Define a modular architecture
• Analyze/Decompose existing solutions in terms of
  this architecture
• Identify existing components that can be
  leveraged in the new modular architecture
• Identify any enhancements required to these
  existing components
• Define new components needed to complete the
  architecture
• Dont reinvent the wheel

8

• Discussion?