Yaakov (J) Stein

1
IRTF Timing RG
TICTOC - IETF - 73 17 Nov. 2008

• Yaakov (J) Stein

2
Charter language
The issue

• our charter says the following
• It is also expected that active individuals in
  the TICTOC WG
• will propose the formation of an IRTF RG to study
  more
• advanced aspects of time and frequency
  distribution.
• there have been no discussions on topics for such
  a RG
• We could call it - Sync Your Network Clock
• In the following slides I will mention possible
  topics of interest

3
Non-goals

• We would probably not want to tackle
• BCP(s) about design of particular algorithms (IPR
  issues)
• comparison of particular algorithms in various
  scenarios
• things already being carried out by TICTOC
• characterization / test scenarios (being done in
  ITU SG15)

4
PDV and clock recovery performance

• The most common question asked by nonexperts is
• given a PDV how good will my clock recovery be ?
• The truth is that this question is not well
  enough defined
• Even if the network is stationary (no reroutes or
  load changes)
• a single number PDV is insufficient to predict
• In general we need the full PDV distribution and
  its full spectrum
• However, in certain cases the question is
  answerable
• In the Gaussian, IID, white-noise case the answer
  is easily obtained (assuming a PLL-like
  mechanism)
• For other cases of interest, simulation using
  existing algorithms
• can provide guidelines

5
PDV characterization

• PDV is usually defined with certain applications
  in mind
• for example - sizing of playout buffers for
  real-time media
• The usual percentile-based definition reflects
  the intended use
• RFC 3393 discusses the PDV's median PDF/CDF
• but not its spectrum
• In addition, RFC 3393 discusses having locked
  clocks
• It would be useful to specify how and what to
  measure
• (coordination with ITU, to avoid duplication,
  will be needed)

6
Time as a commodity

• There is speculation that ubiquitous high-quality
  time
• may transform application and protocol design
• In many cases this has not been sufficiently
  explored
• IxTF documentation of the implications would be
  useful
• In particular, it would be useful to research
• performance measurement when network elements
  have time
• interaction between routing and timing
• interaction between DNS and timing
• database cache coherency
• others ?

7
Applications to security

• There may be improved security mechanisms that
  rely on high-quality time at various points in
  the network
• Public-key cryptography and Autokey mechanisms
  are presently used to secure timing streams,
• but perhaps are not ideal - are there better
  mechanisms ?

8
New timing distribution topologies

• It has been suggestions in the literature that
  time can be distributed in non-hierarchical
  fashion
• (e.g. classless time distribution)
• This means that we need not know a trusted server
  a-priori
• It is unclear if this can be extended to
  frequency distribution due to the problem of
  timing loops
• There may be other novel non-client/server
  topologies

9
Other suggestions

• high-quality time in mobile devices
• improving time in existing IP protocols (ICMP, )
• distributing time over networks with very long
  delay