Time Synchronization

1
Time Synchronization

• Process and Time Synchronization
• Regulation of Local Clock
• Global Clock Synchronization
• Clock Synchronization Algorithms

2
Process and Timing Synchronization

• Distributed systems gt cooperative processes
  connected by a network
• Applications
• A task may consist of multiple events (as
  specified in program logics) to be performed by
  multiple processes (or sample process) in a
  predefined sequence (ordered)
• Execution of a process gt serving an event
• Some events can be executed concurrently and some
  may be performed sequentially gt concurrent
  processes and sequential processes
• Sequential processes have pre-defined precedence
  relationships
• Step 1 input x step 2 input y step 3 z x
  3y
• Can step 1 performs before step 2?
• Note in here, we use the term process which can
  be implemented as a process or a thread

3
Process and Timing Synchronization

• Process state gt the current status of the
  process (i.e., active or blocked, program counter
  value) together with the values of its accessed
  variables
• An event transforms a process from state S to
  state S
• State transition State(active) gt State(wait)
• I.e., The event of sending a message by process P
  changes the state of P from Not yet send to
  sent
• Definition e -gt e event e is occurred before
  event e (precedence relationship)
• When all the events have completed, it is
  expected that the system arrives the expected
  (correct) state gt output
• History the history of a process Pi is a series
  of events that have happened within it until the
  current time
• History(Pi) hi lt ei,1, ei,2, , gt
• How to order the events in a distributed
  environment to ensure that the events occurred
  (served) following the requirements of the
  application/specifications?
• (1) based on the process state gt require IPC
• (2) based on the real time (i.e., current time)gt
  require a global time (time synchronization)

4
Process Synchronization
Different sub-processes may be processed at
different sites The results from a sub-process
may need to return to its parents The sub-process
may be coordinated in processing
T1
T1,1
Precedence relationship
5
Process and Timing Synchronization

• Local clock and global clock
• Lack of a global clock in a distributed system
  although each computer has its own local clock
• Ordering of events at a single site gt by a local
  clock
• The local clock can also be used to time-stamp
  the occur/finish time of an event
• How to order events occurring in different
  processes?
• Process P1 sends a variable v1 message to Process
  P2 and asks P2 add its value v2 to v1, and then
  return the new value after 5 sec of the
  transmission from P1
• So, at which the time P2 should return the
  updated value of v1 to P1?
• The local clock of P1 and P2 may be different
• The communication delay is totally unpredictable
• Clock synchronization gt agree to the same
  absolute time
• In addition, process management also requires an
  accurate global clock
• I.e., concurrency control based on timestamp
  ordering
• Real-time events need to respond to the real
  world event at a specific time. I.e., Once an
  intruder is detected, an action needs to be
  generated within 5 sec

6
Events and Timing

• Relative time Vs. absolute time
• Relative time mainly for event ordering. The
  occur times of events needs to follow the
  specified precedence relationship. I.e., One
  event has to be started only after the finish of
  another event
• Get a Get b Calculate c a b
• Absolute time an event has to be
  started/finished after/before a specific agreed
  global time. I.e., when you press the return
  key, a result needs to be generated within 5 sec
• I.e., in a real-time system, the computer system
  is responding to real-time events (responsive).
  The start time and completion time of a task need
  to follow the real-time constraints of the
  requirements of the real world events
• It could be a relative time constraint, i.e.,
  generate a response after 5 sec of finishing a
  the previous task
• It could be an absolute time constraint, i.e.,
  generate a result on position of an air-craft
  every 5 sec
• What will be the consequence of failing to meet
  the timing requirement of a real-time process???

7
Sources of Accurate Timing Signals

• How to determine and meet the ordering
  requirements???
• We (not just computers) need a highly accurate
  (reliable) global clock for references and
  coordination
• International Atomic Time (TAI) is based on very
  accurate physical clocks (drift rate 10-13)
  (highly accurate quartz clock is 10-8)
• What is draft rate? Clocks are normally not
  advanced in a constant speed
• No perfect clock gt only a more accurate clock
• Coordinated Universal Time (UTC) is an
  international standard for time keeping
• How to define a second?
• One year gt 365 x 24 x 3600sec
• One day gt 24 x 3600 sec
• It is broadcast from radio stations on land and
  satellite (e.g. GPS)
• Computers with receivers can synchronize their
  clocks with these timing signals
• Signals from land-based stations are accurate to
  about 0.1-10 millisecond (errors in estimating
  the delay in transmission)
• Signals from GPS are accurate to about 1
  microsecond

8
Definition of Time
Fr. Tanenbaum
Computation of the mean solar day the interval
between two consecutive transits of the sun
(highest point in the sky) is called the solar day
9
Physical Clock
Fr. Tanenbaum
Solar second vs. TAI second
TAI seconds are of constant length, unlike solar
seconds. Leap seconds are introduced when
necessary to keep in phase with the sun.
10
Regulation of Local Clock

• Each computer has a hardware clock Hi(t)
• A computer timer is a precise quartz crystal that
  oscillates at a fixed frequency (i.e., n
  oscillations one TAI sec)
• Timer generates a clock tick (an interrupt) after
  a fixed number of oscillates
• Note the time interval between successive events
  must be larger than the clock resolution (the
  period of clock tick)
• For process pi
• Ci(t) software clock (local clock)
• Hi(t) hardware clock, I.e., the time given by
  hardware clock
• Ci(t) ?H(t) ? (a conversion of the hardware
  clock)
• In real cases, Ci(t) cannot be the same as t (TAI
  value) all the times due to clock drift
  (imperfect)
• Clock drift different clocks count time at a
  slightly different way (the oscillation is not
  perfect
• It is common to define a hardware clock H to be
  correct if its drifts fall within a known bound ?
  gt 0 (e.g. for real-time t and t, tgtt)
• (1- ? )(t-t) H(t) H(t) (1? )(t-t)
• Monotonicity t gt t gt C(t) gt C(t)

11
Regulation of Local Clock
What is wrong with the figure?
The clock rate may not be a constant
Fr. Tanenbaum
The relation between clock time and UTC when
clocks tick at different rates.
12
Clock Synchronization

• Clock synchronization to synchronize the clocks
  at different computers to agree at the same
  (similar) reading (i.e., close to the UTC time)
• How to synchronize your watches?
• Difficulties in clock synchronization in a
  distributed environment
• Not all sites have direct access to accurate time
  source such as a GPS receivers due to the cost
• Sites has to synchronize their local clocks with
  those have more accurate time (a common reference
  clock)
• How to determine which one is more accurate?
  Where?
• Synchronization needs to be done periodically due
  to clock drift rate is change in the offset
  (difference in readings) between the clock and a
  normal perfect reference clock per unit of time
  measured (about 10-6 or 1 sec. in 11.6 days)
• How to determine the synchronization period?
  Based on the value of ?
• If a sites clock is ahead of the reference (time
  server) to which it synchronizes to, it cannot
  be simply set back (why?)
• This may have the effect that an event that has
  happened but its time-stamp is in the future
• Slow it down until it reaches the desire value
  (UTC value)

13
Skew between Computer Clocks in a Distributed
System
Reliable clock
Periodically broadcast clock readings
Fr. Dollimore
Different clocks may have different oscillate
periods and advancement rate. This rate also
change slightly with the environmental conditions
14
External Internal Clock Synchronization

• External synchronization
• A computers clock Ci is synchronized with an
  external authoritative time source S (a reliable
  source), so that
• S(t) - Ci(t) lt D for i 1, 2, N over an
  interval, I of real time
• The clocks Ci are accurate to within the bound D
• Internal synchronization
• The clocks of a pair (group) of computers are
  synchronized with other
• Ci(t) - Cj(t) lt D for i 1, 2, N over an
  interval, I of real time
• The clocks Ci and Cj agree within the bound D
• Good for relative events synchronization (what?
  Relative clocks)
• Internally synchronized clocks are not
  necessarily externally synchronized, as they may
  drift collectively (all are not accurate)
• If the set of processes P is synchronized
  externally within a bound D, it is also
  internally synchronized within bound 2D ( D)
• Synchronization problem estimation of the delay
  in communication
• The delay is always quite vary in external
  synchronization (why?)
• How about the case for internal synchronization?

15
Clock Synchronization Cristian Algorithm

• No upper bound on message transmission delay but
  it has a lower bound
• The Cristian algorithm for systems with short
  transmission delays
• Client C requests time in Request message sent to
  time server S, then it receives time value t in
  message CUTC from S
• S is connected to an accurate time reference
• t is the current time in S before sending back
  CUTC to C
• Let Ttrans be time taken for CUTC message from S
  to C, then C should set its time to t Ttrans
• Ttrans can be variant. You may say Ttrans
  minx, x 0 and min time of message
  transmission if no interference of other process
  and no other messages, but x is unknown
• Solution
• (1) Record total round trip time as Tround
  (between sending Request and receiving CUTC,,
  i.e., T1-T0)
• (2) If the time received the Request message is
  t, then C can estimate its time as t Tround/2
• Counting multiple requests to get the average
• What is the main assumption?

16
Clock synchronization using a time server
17
Cristian's Algorithm
t
t
Fr. Tanenbaum

• Getting the current time from a time server

18
Clock Synchronization Cristian Algorithm

• When C receives CUTC, Cs time is t Tround/2
• Let the minimum transmission time be min, and min
  is known
• Therefore, t (the time in S when C receives CUTC
  ) lies in the range tmin, tTround - min
• t min the earliest time that S receives the
  request
• t Tround min the latest time that S sends
  CUTC
• Then, t Tround -min - (tmin) Tround - 2min
• The accuracy for this time adjustment is
  ?(Tround/2 - min)
• Problem single-server failure
• Solution providing group synchronization time
  servers (tradeoffs?)

19
Clock Synchronization Berkeley Algorithm

• Berkeley Algorithm for time (internal)
  synchronization
• In Berkeley UNIX, one of a group sites
  (computers) is chosen as coordinator, (master or
  time server)
• (1) The master (actually a time daemon in the
  master) periodically polls the other sites
  (slaves) to ask what time it is there, by telling
  its own time (the time in master) to these slaves
  
• (2) The slaves response with how far ahead or
  behind the time daemon they are, that is, the
  difference between its time and the time in
  master
• (3) Based on the replies from the slaves, the
  master averages the time obtained (including its
  own)
• (4) The master sends time rate adjust value ( or
  -) to slaves, requesting them to adjust their
  time rate
• Question How does the Cristians algorithm be
  applied to the above Berkeley algorithm?
• How about if the master is failed? Find another
  master? How?

20
The Berkeley Algorithm An Example
Fr. Tanenbaum

• The time daemon asks all the other machines for
  their clock values
• The machines answer
• The time daemon tells everyone how to adjust
  their clock

21
Clock Synchronization Network Time Protocol
(NTP)

• NTP is used in the Internet (a widely distributed
  system with great variability in communication
  delays)
• Why do we need NTP instead of using the Cristian
  algorithm? Can the Cristian algorithm be applied
  for the Internet?
• A statistical techniques for filtering timing
  data with adaptation to connection
• Time servers are arranged in levels called
  strata. NTP intends to
• Provide a service enabling clients in Internet to
  be synchronized accurately to UTC
• Provide reliable service to losses of
  connectivity
• enable clients to resynchronize sufficiently and
  frequently
• Provide protection against interference with the
  time service

22
Clock Synchronization Network Time Protocol

• Synchronization clock Network Time Protocol
  (NTP)
• Stratum 1 Primary server (PS) sync directly with
  UTC sources.
• Stratum 2 Secondary servers (SS) sync directly
  to PS.
• Stratum 3 Lowest servers (LS) execute in user
  sites sync with SS.
• Accuracy the number of levels (strata)

Less accurate
23
Clock Synchronization Network Time Protocol

• Multiple servers provide for redundancies and
  hence availability of time sources and
  fault-tolerance
• NTP synchronizes with one another in one of three
  modes
• Multicast Used on LAN. One or more servers
  periodically broadcast the time to servers
  running in workstations which set their clocks
  assuming a small delay. Accuracy is low
• Procedure-call Similar to Cristians method.
  Workstations ask the server the time. Accuracy is
  higher than multicast
• Symmetric mode Used by high level to achieve
  highest accuracy. A pair of servers exchange
  messages bearing timing information

24
Clock Synchronization Network Time Protocol

• Messages are delivered unreliably using UDP (why
  not TCP?)
• Each message bears the timestamps of recent NTP
  messages
• The local time when the previous NTP message
  between the pair was sent and received
• The local time when the current message was sent
• For each pair of messages sent, calculate an
  offset oi to estimate the actual offset o between
  the two clocks, and a delay di the total
  transmission delay of the two messages
• What is the difference between o and oi?
• o tb - ta, where tb is the time at server B
  when ta is the time at server A
• Taking the communication delay
• Ti-2 Ti-3 t o and Ti Ti-1 t o
• The total communication delay
• di t t Ti-2 - Ti-3 Ti - Ti-1

25
Clock Synchronization Network Time Protocol

• The estimated offset oi
• Since oi is to estimate tb - ta, oi ((Ti-2 -
  Ti-3) - t (Ti-1 - Ti) t)/2
• oi (Ti-2 - Ti-3 Ti-1 - Ti) /2 assuming that t
  is similar to t
• o Ti-2 - Ti-3 - t
• o Ti-1 - Ti t
• Thus, o (Ti-2 - Ti-3 - t Ti-1 - Ti t) / 2
  (Ti-2 - Ti-3 Ti-1 - Ti t - t ) / 2
• o oi (t t) /2
• Since t, t gt 0, oi - di/2 lt o lt oi di/2
  (case 1 t 0, case 2 t 0)
• Thus, oi is an estimate of the offset while di is
  the accuracy of the estimate
• We can associate a pair lt oi, digt for each
  connection
• Apply a data filtering algorithm to successive
  pairs lt oi, digt of to estimate the offset o and
  calculates the quality of this estimate called
  filter dispersion

26
Messages exchanged between a pair of NTP peers
t
t
27
Clock Synchronization Network Time Protocol

• A relatively high dispersion of successive pairs
  represents relatively unreliable data. A
  reconfiguration may be needed
• The eight most recent pair lt oi, digt are
  retained. The value oj that corresponding to the
  minimum value dj is chosen to estimate o

28
References

• Dollimore ch 11.1 to 11.3
• Tanenbaum ch 5.1