[T IME ] Shrideep Pallickara Computer Science Colorado State - - PDF document

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.1

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS 555: DISTRIBUTED SYSTEMS

[TIME]

Shrideep Pallickara Computer Science Colorado State University

August 29, 2019

L2.1 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.2 Professor: SHRIDEEP PALLICKARA

Frequently asked questions from the previous class survey

August 29, 2019

¨ Quizzes/Exams? ¤ Frequency, etc.

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.2

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.3 Professor: SHRIDEEP PALLICKARA

Topics covered in this lecture

August 29, 2019

¨ Physical Clocks ¨ How time is actually measured ¨ Clock synchronization algorithms ¤ Network Time Protocol

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

SYNCHRONIZATION

August 29, 2019

L2.4

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.3

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.5 Professor: SHRIDEEP PALLICKARA

Synchronization

August 29, 2019

¨ Important that processes do not access a shared resource

simultaneously

¤ Cooperate in granting each other temporary, exclusive access ¨ Multiple processes must sometimes need to agree on ordering of

events

¤ Was m1 from process P sent out before or after m2 from process Q ?

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.6 Professor: SHRIDEEP PALLICKARA

What we will look at:

August 29, 2019

¨ Synchronization based on actual time ¨ Synchronization in which only relative ordering matters ¤ Rather than ordering in absolute time

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.4

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.7 Professor: SHRIDEEP PALLICKARA

Let’s look at a make example

August 29, 2019

¨ Large programs split up into multiple source files ¨ make examines times at which all source and object files were

modified

¨ input.c (@ 2151) and input.o (@ 2150)? ¤ Input.c has changed since input.o was created ¨ output.c (@2144) and output.o (@2145)? ¤ No recompilation required

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.8 Professor: SHRIDEEP PALLICKARA

What could happen in a distributed system with make?

August 29, 2019

¨ output.o (@2145) ¨ output.c modified right after that ¤ But assigned (@2144) by the machine ¨ No recompilation required ¤ Incorrect! ¤ Executable is a mix of object files from new and old sources

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.5

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

PHYSICAL CLOCKS

August 29, 2019

L2.9 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.10 Professor: SHRIDEEP PALLICKARA

Nearly all computers have a circuitry for tracking time

August 29, 2019

¨ Timer is a more precise term than clock ¨ Quartz crystals are kept under tension ¤ Oscillates at a well-defined frequency ¤ Frequency depends on: n Type and cut of crystal n The amount of tension

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.6

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.11 Professor: SHRIDEEP PALLICKARA

Each crystal has two registers associated with it

August 29, 2019

¨ {Counter, Holding Register} ¨ Each oscillation decrements the counter ¨ When counter reduces to zero ¤ Interrupt is generated: clock tick ¤ Counter reloaded from the holding register ¨ We can program the timer to generate interrupts N times per second

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.12 Professor: SHRIDEEP PALLICKARA

Measuring time

August 29, 2019

¨ Time is stored as number of ticks ¤ After some known starting date and time ¨ Stored in special battery-backed CMOS RAM

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.7

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.13 Professor: SHRIDEEP PALLICKARA

How the OS uses the timer to report actual time

August 29, 2019

¨ The OS reads the node’s hardware timer, Hi(t), scales it and adds an

• ffset

§ Ci(t) = αHi(t) + β ¤ This is the software clock ¨ In general, the clock is not completely accurate § Ci(t) will always differ from t

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.14 Professor: SHRIDEEP PALLICKARA

On a single machine it is ok if the clock is off by a small amount

August 29, 2019

¨ All processes use the same clock ¨ Processes are internally consistent ¨ Only relative times matter in some cases

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.8

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.15 Professor: SHRIDEEP PALLICKARA

Things change when we add multiple clocks

August 29, 2019

¨ Frequency at which the quartz oscillator runs is fairly stable ¨ Impossible to guarantee that all oscillations are at the same frequency

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.16 Professor: SHRIDEEP PALLICKARA

This causes the (software) clocks to slowly get out of synch

August 29, 2019

¨ You get different values during read outs ¤ Clock skew ¨ Time is not independent of the machine on which it was reported

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.9

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.17 Professor: SHRIDEEP PALLICKARA

Physical and software clocks

August 29, 2019

¨ Multiple physical clocks are desirable ¤ Efficiency ¤ Redundancy ¨ How do we synchronize software clocks with: ¤ Real-world clocks ¤ Each other

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.18 Professor: SHRIDEEP PALLICKARA

Clock skew and drift

August 29, 2019

¨ The instantaneous difference between the readings of any two clocks is

called their skew

¨ All clocks (including crystal-based ones) are subject to clock drift ¤ This means that they count time at different rates

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.10

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.19 Professor: SHRIDEEP PALLICKARA

Clock drift rates

August 29, 2019

¨ The drift rate is the change in the offset between the clock and a

nominal perfect reference clock per unit of time

¤ Offset: Difference in reading ¨ Some typical drift rates: ¤ Regular Clocks based on quartz crystal n 10-6 seconds/second n i.e., 1 second every 1,000,000 seconds or 11.6 days ¤ High precision quartz clocks: 10-7 seconds/second ¤ Atomic clocks: 10-13 seconds/second

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

HOW TIME IS ACTUALLY MEASURED

August 29, 2019

L2.20

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.11

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.21 Professor: SHRIDEEP PALLICKARA

Since the invention of mechanical clocks, time has been measured astronomically

August 29, 2019

¨ Sun’s highest apparent position in the sky ¤ Transit of the sun ¤ At about noon each day ¨ Interval between consecutive transits of the sun ¤ Solar day ¨ Solar second? ¤ Solar day/86400

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.22 Professor: SHRIDEEP PALLICKARA

But the earth’s rotation is not constant

August 29, 2019

¨ Proved in the 1940s ¨ Earth is slowing down ¤ Tidal friction ¤ Atmospheric drag ¨ Based on growth patterns on ancient corals ¤ 300 million years ago, we had 400 days per year

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.12

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.23 Professor: SHRIDEEP PALLICKARA

The length of the year is not believed to have changed

August 29, 2019

¨ Days have simply gotten longer ¤ Long-term trend ¨ Short term variations ¤ Due to turbulence in the earth’s core ¨ Astronomers measure a large number of days ¤ Take their average ¤ And then divide by 86400

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.24 Professor: SHRIDEEP PALLICKARA

Physicists have taken over the job of timekeeping from the astronomers

August 29, 2019

¨ Atomic clock invented in 1948 ¨ Measures time much more accurately ¤ Independent of the wiggling and wobbling of earth ¨ Count energy transitions of the cesium 133 atom

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.13

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.25 Professor: SHRIDEEP PALLICKARA

Atomic seconds

August 29, 2019

¨ 1 second is the time for a cesium 133 atom to make 9,192,631,770

transitions

¤ Transitions are between two hyperfine levels of the ground state of cesium-

133 (Cs133)

¨ Why this number? ¤ Atomic second = mean solar second in the year of its introduction

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.26 Professor: SHRIDEEP PALLICKARA

Several labs around the world have cesium 133 atomic clocks

August 29, 2019

¨ Periodically, each lab tells the Bureau International de L’Heure (BIH) in

Paris about their ticks

¨ BIH averages this to produce TAI ¤ International Atomic Time ¨ TAI is mean number of ticks of cesium-133 ¤ Since Jan 1, 1958 divided by 9,192,631,770

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.14

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.27 Professor: SHRIDEEP PALLICKARA

TAI is highly stable, but …

August 29, 2019

¨ 86400 TAI seconds is now slower than the mean solar day ¤ Days are getting longer ¨ Using TAI exclusively means that over the years noon would get earlier

and earlier …

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.28 Professor: SHRIDEEP PALLICKARA

People notice these kind of things ...

August 29, 2019

¨ In 1582, Pope Gregory XIII decreed 10 days be removed from the

calendar

¨ Led to riots! ¤ Landlords demanded full month’s rent ¤ Bankers wanted full month’s interest ¤ But employers refused to pay for work that was not done ¨ Protestant countries did not accept the Gregorian calendar for 170

years

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.15

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.29 Professor: SHRIDEEP PALLICKARA

BIH solves this problem by introducing leap seconds

August 29, 2019

¨ Whenever the discrepancy grows to 800 milliseconds ¤ Total number of leap seconds introduced so far? 30 ¨ Time based on TAI seconds; but in phase with apparent motion of the

sun

¤ This is the Universal Coordinated Time (UTC) ¨ UTC has replaced the Greenwich Mean Time

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.30 Professor: SHRIDEEP PALLICKARA

UTC uses and availability

August 29, 2019

¨ Power companies synchronize timing of their 50/60Hz clocks to UTC ¨ NIST operates a shortwave radio station in Ft Collins ¤ Call letters WWV (accuracy of WWV is ± 1 msec) ¤ Accuracy is ± 10 msec due to atmospheric fluctuations ¨ Radio receivers for WWV and other UTC sources

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.16

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

GLOBAL POSITING SYSTEM

August 29, 2019

L2.31 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.32 Professor: SHRIDEEP PALLICKARA

GPS

¨ Uses 30 satellites each orbiting at a height of approximately 20,000

km

¨ Each satellite has up to four atomic clocks ¤ Regularly calibrated from special stations on Earth

August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.17

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.33 Professor: SHRIDEEP PALLICKARA

Broadcasting

¨ Each satellite continuously broadcasts its position ¤ Timestamps each message with its local time ¨ Broadcasting allows every receiver on Earth to accurately compute its

• wn position

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.34 Professor: SHRIDEEP PALLICKARA

Calculating the receiver’s position [1/4]

¨ Let Δr denote the deviation of the receiver’s clock from actual time ¨ When a message is received from satellite i with timestamp Ti ¤ Measured delay Δi by the receiver consists of two components

• Δi = (Tnow- Ti) + Δr

August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.18

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.35 Professor: SHRIDEEP PALLICKARA

Calculating the receiver’s position [2/4]

¨ Signals travel at the speed of light, c ¨ Measured distance of the satellite is cΔi ¨ If the receiver’s position is (xr, yr, zr) ¤ The real distance is also

di = (xi − xr)2 + (yi − yr)2 + (zi − zr)2

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.36 Professor: SHRIDEEP PALLICKARA

Calculating the receiver’s position [3/4]

¨ When we have 4 satellites ¤ We have 4 equations in 4 unknowns ¤ Allows us to solve the coordinates (xr, yr, zr) and Δr

August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.19

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.37 Professor: SHRIDEEP PALLICKARA

Calculating the receiver’s position [4/4]

¨ GPS does not take leap seconds into account ¤ Such an error can easily be compensated for in hardware ¨ Other sources of error: ¤ Atomic clocks in satellites are not in perfect sync ¤ Receiver’s clock has finite accuracy ¤ Signal propagation speed is not constant n Signals slow down when entering e.g., the ionosphere

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.38 Professor: SHRIDEEP PALLICKARA

Accuracy of GPS receivers

¨ Commodity GPS receivers can be precise within a range of 1-5 meters ¨ Professional receivers have a claimed error of less than 20-35

nanoseconds

August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.20

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CLOCK SYNCHRONIZATION ALGORITHMS

August 29, 2019

L2.39 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.40 Professor: SHRIDEEP PALLICKARA

Goals of Synchronization

¨ If 1 machine has a WWV receiver ¤ Keep all machines synchronized to it ¨ If no machines have a WWV receiver ¤ Keep all machines in synch as well as possible

August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.21

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.41 Professor: SHRIDEEP PALLICKARA

Modeling the setting [1/2]

¨ Timer causes an interrupt H times per second ¨ When timer goes off, interrupt handler adds 1 to software clock ¤ Tracks ticks from some agreed-upon time ¤ Value of this clock, C ¨ UTC time, t

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.42 Professor: SHRIDEEP PALLICKARA

Modeling the setting [2/2]

¨ When UTC time is t ¨ Value of clock on machine p is Cp(t) ¨ In a perfect world ¤ Cp(t) = t for all p

Cp

' (t) = dC

dt =1 Frequency of p’s clock at time t Cp

' (t) August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.22

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.43 Professor: SHRIDEEP PALLICKARA

Skews and Offsets

¨ Offset relative to a specific time t is Cp(t) – t ¨ Skew indicates extent to which frequency differs from that of a

perfect clock Cp

' (t) −1

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.44 Professor: SHRIDEEP PALLICKARA

Timer specifications

August 29, 2019

¨ Real timers do not interrupt exactly H times per second ¨ With H = 60; a timer should generate 216000 ticks per hour ¨ Relative error in modern timer chips is about 10-5 ¤ In our example of H=60; about ±2 ¤ Between 215,998 to 216,002 ticks per hour ¤ Manufactures usually specify a drift rate

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.23

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.45 Professor: SHRIDEEP PALLICKARA

Clocks ticking at different rates

Clock Time, C UTC, t dC dt =1

P e r f e c t C l

• c

k Slow Clock

dC dt <1

Fast Clock

dC dt >1

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

NETWORK TIME PROTOCOL

August 29, 2019

L2.46

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.24

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.47 Professor: SHRIDEEP PALLICKARA

Network Time Protocol

¨ Clients connect to a time server ¨ Time server provides accurate time ¤ Has WWV receiver ¤ Accurate clock ¨ CHALLENGE ¤ Network latencies ¤ Time is outdated by the time you get it

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.48 Professor: SHRIDEEP PALLICKARA

Getting time from a time server

B A

TIME SERVER CLIENT

T2 T3 T1 T4 Offset at A = T3 – T4 + time the message was in transit Offset at A = T3 – T4 + (T2 – T1) + (T4 – T3)

2

Delay = (T4 – T1) + (T3 – T2)

2

August 29, 2019

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.25

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.49 Professor: SHRIDEEP PALLICKARA

Let’s look at an example

B A

TIME SERVER CLIENT

T1=10 T4=25

100 T2=115 T3=120

Offset at A = T3 – T4 + (T2 – T1) + (T4 – T3)

2

= 120 – 25 + (115 – 10) + (25 – 120)

2

= 100

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.50 Professor: SHRIDEEP PALLICKARA

In the NTP protocol, both the offsets and delays are used

August 29, 2019

¨ Eight pairs of {offset, delay} tuples are buffered ¨ Minimum value of delay is chosen ¤ The corresponding offset is taken as the most reliable one

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.26

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.51 Professor: SHRIDEEP PALLICKARA

Time adjustment rules

¨ Time is not allowed to run backward ¤ Even though the client’s clock is fast ¨ Changes are introduced gradually ¤ If timer is set to generate 100 interrupts/second ¤ Each interrupt adds 10 milliseconds to the clock n To slow down we would add only 9 milliseconds n To speed up we would add 11 milliseconds

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.52 Professor: SHRIDEEP PALLICKARA

It is possible to apply NTP symmetrically

August 29, 2019

¨ If time server is known to be more accurate ¤ No point adjusting time with the client ¨ NTP divides servers into strata ¨ Server with a WWV receiver or atomic clock ¤ Stratum-1 server (clock is stratum-0)

SLIDES CREATED BY: SHRIDEEP PALLICKARA L2.27

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.53 Professor: SHRIDEEP PALLICKARA

Communication and changing stratums

¨ A contacts B for offset adjustments ¤ Only if A’s stratum is higher than B ¤ E.g. A may be stratum 5, and B is stratum-1 ¨ After offset adjustments ¤ If B was a stratum-k, A becomes stratum-(k+1) ¤ E.g. A is stratum-5, B is stratum-1 n Then A becomes stratum-2

August 29, 2019 CS555: Distributed Systems [Fall 2019]

• Dept. Of Computer Science, Colorado State University

L2.54 Professor: SHRIDEEP PALLICKARA

The contents of this slide-set are based on the following references

August 29, 2019 ¨ Distributed Systems: Principles and Paradigms. Andrew S. Tanenbaum and Maarten Van

der Steen. 2nd Edition. Prentice Hall. ISBN: 0132392275/978-0132392273.

¨ Wikipedia: Atomic Clocks http://en.wikipedia.org/wiki/Atomic_clock