Explain the difference between logical and physical clocks.

Physical clocks measure the time of day. Logical clocks

are used to mark relationships among events in a
distributed system.
Selected Questions

Exam 2
Spring 2008

Page 1 Page 2

What problem with Lamport clocks to vector clocks solve? A client's clock reads 3:20:00. The server's clock reads 3:10:00
when they synchronize using the Berkeley algorithm. Assume
message delays are negligible. What is the time at the client after
synchronization?
With Lamport clocks, you cannot tell whether two
events are causally related or concurrent by looking at
the timestamps. Just because L(a) < L(b) does not mean The Berkeley algorithm averages clocks among the
that a->b. entire group. In this case, the group has two members:
the client and the server. The average of the two clocks
is (3:20:00 + 3:10:00)/2 = 3:15. Both the client and
Vector clocks allow you to compare two vector
server will be set to 3:15.
timestamps to determine whether the events are
concurrent or not.

Page 3 Page 4

A client's clock reads 3:20:00. The server's clock reads 3:10:00 Some networks, such as cable TV Internet service, provide an asymmetric
when they synchronize using Cristian's algorithm. Assume message bandwidth. For example, a cable modem may provide 12 Mbps downstream
delays are negligible. What is the time at the client after service but only 1 Mbps upstream. Cristian's algorithm assumes symmetric
synchronization? delays to and from the server.
(a) Reformulate the algorithm to accommodate asymmetric delays, where
TS is the server's time, Ta is the time the request was sent, Tb is the time
the response was received, U is the upstream bandwidth, and D is the
Cristian’s algorithm assumes that the server has an downstream bandwidth.
accurate clock. The client requests the time and sets its
clock to the server’s time + ½(round trip delay). In this Assume the message size is x. The best-case time to send a message is x/
case, the round-trip delay is negligible (i.e., 0), so the U. The best-case time to receive a response is x/D.
The total time is (x/U + x/D) = x(D+U)/DU.
client just sets its clock to the server’s time: 3:10:00.
The ratio of the time to get a response to the total time is:
[x/D] / [x(D+U)/DU] = xDU / x(D+U) = U/D+U. We assume that the
response was generated this long ago compared to the total time.
The special-case Cristian’s algorithm sets the clock to
Tclient = TS + (U/(D+U)) (Tb-Ta)

Page 5 Page 6

1
(b) You are on a network with a 9 Mbps downstream bandwidth and The Coda distributed file system does not support:
a 1 Mbps upstream bandwidth. Your client makes a request at  (a) Tokens for file locking.
9:10:00.0 and gets a response 200 msec later. If the time on the (b) Disconnected clients.
server is 8:42:00.0, what would the time be set to on your client?
(c) Replicated servers.
(d) Session semantics.
From (a):
U = 1 Mbps, D = 9 Mbps
Tclient = TS + (U/(D+U)) (Tb-Ta)
Tclient = TS + (1/10) (Tb-Ta)
= 8:42:00.0 + 0.1 (0:00:00.200)
= 8:42:00.02

Page 7 Page 8

The purpose of DFS tokens is to: SMB is best characterized as a:

 (a) Give the server control over which actions clients can  (a) Remote call, connection-oriented file system.
perform and how they may cache data. (b) Remote call, connectionless file system.
(b) Ensure mutual exclusion is granted when servers are accessed. (c) File upload/download, connectionless file system.
(c) Support read/write replication on servers through token (d) File upload/download, connection-oriented file system.
exchange.
(d) To avoid the overhead of having clients cache file data.

Page 9 Page 10

Opportunistic locks, or oplocks: Alice has a remote file open on her system with a Level 1 oplock.
 (a) Tell the client how it may cache file data and attributes. Bob tries to open the file from another system. The most likely
(b) Are a first-come, first-served mechanism for locking a file. scenario is:
(c) Allow clients to grab locks on a remote file under CIFS.  (a) Alice's oplock will be revoked and Bob will be able to share
the file.
(d) Guarantee exclusive access to a file once it is open.
(b) Bob's request will fail.
(c) Bob will be blocked until Alice releases her lock.
(d) Alice's oplock will be revoked and she will no longer have access
to the file.

Page 11 Page 12

2
A synchronization subnet is: IP multicast uses:
 (a) The collection of NTP servers on the Internet.  (a) Unreliable multicast.
(b) An IP multicast group used for clock synchronization (b) Reliable multicast.
(c) The set of machines addressed by an NTP server operating in (c) Atomic multicast.
multicast mode (d) User-selectable reliability.
(d) The set of NTP servers with which you are currently
synchronizing.

Page 13 Page 14

False sharing in a distributed shared memory system arises A distributed directory in a distributed shared memory system:
because: (a) Ensures cache coherence in a sequentially coherent memory
 (a) Unrelated data is shared between multiple systems but system.
resides on the same memory page. (b) Allows concurrent queries and updates to a memory page.
(b) One distribute memory page is mapped onto two or more (c) Is necessary for systems that employ caching.
systems at the same time.  (d) Distributes the load of looking up page owners.
(c) Two or more distributed memory pages are mapped onto the
same system.
(d) Bad message ordering makes it appear that data is shared when,
in reality, it is not.

Page 15 Page 16

The ring election algorithm works by: Which event is concurrent with the vector clock (2, 8, 4)?
(a) Having all nodes in a ring of processors send a message to a (a) (3, 9, 5)
coordinator who will elect the leader. (b) (3, 8, 4)
(b) Sending a token around a set of nodes. Whoever has the token (c) (1, 7, 3)
is the coordinator.  (d) (4, 8, 2)
(c) Sending a message around all available nodes and choosing the
first one on the resultant list.
 (d) Building a list of all live nodes and choosing the largest
numbered node in the list.

Page 17 Page 18

3
If a DES key is 56 bits, a triple-DES key is: Which statement about the Vigenère ciphers is not true?
(a) 56 bits  (a) It is a block cipher that uses a secret key.
 (b) 112 bits (b) It is a polyalphabetic substitution cipher.
(c) 144 bits (c) It is a collection of Cæsar ciphers.
(d) 224 bits (d) It attempts to be resilient to frequency analysis.

Page 19 Page 20

The Ricart & Agrawala distributed mutual exclusion algorithm is:

(a) More efficient and more fault tolerant than a centralized
algorithm.
(b) More efficient but less fault tolerant than a centralized
algorithm.
(c) Less efficient but more fault tolerant than a centralized
Selected Questions
algorithm.
 (d) Less efficient and less fault tolerant than a centralized
algorithm.

Exam 2
Spring 2008

Page 21 Page 22

The clock in the clock tower in the town of Chronos broke. It was The clock in the clock tower in the town of Chronos broke. It was
repaired but now the clock needs to be set. A train leaves for the repaired but now the clock needs to be set. A train leaves for the
nearest town, Temporis, 100 miles away. It returns 4½ hours later nearest town, Temporis, 100 miles away. It returns 4½ hours later
wth a report that the time according to the clock tower in wth a report that the time according to the clock tower in
Temporis is 4:05. Temporis is 4:05.

(a) To what value should the time be set on the (b) The maxiumum speed that the train can
clock tower? travel is 50 mph. What is the error of the
Cristian’s algorithm: Tserver + ½(round-trip delay)
clock?
Round-trip delay = 4:30 Best case: 100 miles @ 50 mph = 2 hours
4:05 + (4:30/2) = 4:05 + 2:15 = 6:20 Minimum time traveling: 4 hours
Uncertainty window: 4:30 - 4:00 = 30 min
Error = ±15 min

Page 23 Page 24

4
Event a has a Lamport timestamp of 4. Which event is concurrent with the vector
Event b has a Lamport timestamp of 8. timestamp (2, 4, 6) ?
What can we tell about events a and b ?
a)  (3, 5, 7)
a)  Events a and b are causally related. b)  (1, 3, 5)
b)  Events a and b are concurrent. c)  (1, 4, 6)
c)  Event a happened before event b. d)  (1, 4, 7)
d)  If events a and b are causally related, then
event a happened before event b.

Page 25 Page 26

A client has a time of 5:05 and a server has a Which offers the most fault-tolerant message
time of 5:25. Using the Berkeley algorithm, the delivery?
client’s clock will be set to:

a)  Atomic multicast.

a)  5:15 (5:05 + 5:25)/2 = 5:30/2 = 5:15
b)  Totally ordered reliable multicast.
b)  5:20
c)  Causally ordered reliable multicast.
c)  5:25
d)  Hardware multicast.
d)  5:30

Page 27 Page 28

When we had to send an IP packet over an ethernet

network, we found the corresponding ethernet address 28-bit multicast group ID
via the Address Resolution Protocol (ARP). How do we find
the ethernet address that corresponds to a particular IP Lower 23 bits
multicast address? 32-bit class D IP address:

a)  All IP multicast packets use the same reserved 1110 XXXX Xxxx xxxx xxxx xxxx xxxx xxxx

ethernet address.
48-bit ethernet address: Multicast bit
b)  ARP is also responsible for handling multicast
addresses. 0000 0001 0000 0000 0101 1110 0xxx xxxx xxxx xxxx xxxx xxxx

c)  By mapping some of the bits of an IP multicast

address onto an ethernet address. 23-bit vendor component

d)  By mapping all of the bits of an IP mulicast address

onto an ethernet address.

Page 29 Page 30

5
How does IP multicast achieve reliable packet A copyset is used:
delivery?

a)  In a DSM system that employs page caching.

a)  It doesn’t
b)  In a DSM system as an alternative to page
b)  Via best-effort multicast.
caching.
c)  Via atomic multicast.
c)  To implement lazy release consistency.
d)  Via IGMP (Internet Group Management
d)  To implement entry consistency.
Protocol)

Page 31 Page 32

An ethernet card may support all of the If a frequency analysis of ciphertext reveals a
following mechanisms for multicast except: statistical equivalence to the corresponding
frequencies of plain English text, we can guess
a)  Filtering packets based on a hash of the that the cipher is:
multicast address.
a)  a monoalphabetic substitution cipher.
b)  Searching through a small table for a
b)  a polyalphabetic substitution cipher.
matching ethernet address.
c)  a transposition cipher.
c)  Searching through a larger table of ethernet
addresses in main memory. d)  a rotor machine.
d)  Accepting all multicast packets via multicast No substitution of characters takes place in
promiscuous mode. a transposition cipher. If the plaintext had
2,900 E’s and 310 Z’s, so will the ciphertext.
Page 33 Page 34

A rotor machine implements: False sharing in a DSM system refers to:

a)  a monoalphabetic substitution cipher. a)  an unwanted page being cached on another
b)  a polyalphabetic substitution cipher. system
c)  a transposition cipher. b)  unrelated data being resident on the same
d)  a one-time pad. page.
c)  invalid data in the directory, leading the
system to falsely believe the page is shared.
d)  two versions of the same page being present
on one system.

Page 35 Page 36

6
A bully election algorithm: Which mutual exclusion algorithm works wen the
membership of the group is unknown?

a)  picks the first process to respond to an

election request. a)  Centralized.
b)  relies on majority vote to pick the winning b)  Ricart-Agrawala.
process. c)  Lamport.
c)  assigns the role of coordinator to the d)  Token Ring.
processs holding the token at the time of
election.
d)  picks the process with the largest ID.

Page 37 Page 38

Weak consistency models offer an advantage

over sequential consistency because:

a)  memory synchronization does not have to

occur with each memory operation. Selected Questions
b)  caching may be used.
c)  only the needed data needs to be shared, not Spring 2006
Exam 2
the entire memory system.
d)  they ensure that each write operation
invalidates or updates all cached copies
before the next instruction is executed.
Page 39 Page 40

What is false sharing? Assign Lamport timestamps to the following events:

•  When processes are accessing unrelated data that 5 a 6 b 7

happens to reside on the same shared page. This P0
leads to thrashing.
1 e
P1
c 6 d 7 8
•  Not: when different data resides on the same page. 1
P2
f 9

a. 6 d. 7

b. 7 e. 8

c. 6 f. 9

Page 41 Page 42

7
You have several computers cooperating on a parallel
11. Which set of events is concurrent (all events are
program. CPU 1 tries to write to page 302, which is
concurrent with each other)?
currently owned by CPU 0.
Explain the sequence of operations that takes place in a.  (3, 1, 5, 7), (3, 2, 6, 7), (2, 1, 6, 8)
a DSM system. b.  (2, 1, 3, 4), (2, 2, 3, 3), (3, 3, 2, 5)
c.  (1, 2, 3, 4), (2, 3, 4, 5), (3, 4, 5, 6)
•  CPU 1 attempts to write and get a page fault. d.  (1, 5, 6, 7), (1, 4, 5, 7), (1, 3, 2, 2)
•  Page fault handler contacts directory to find current owner of
page 302. If we do an element-by element comparison:
•  CPU 1 fetches page from CPU 0 (CPU 0 sends page). (2, 1, 3, 4) is neither ≥ nor ≤ to (2,2,3,3)
•  CPU 0 creates a free page frame and loads page 302 into it. 2=2; 1<2; 3=3, 4>3.
•  CPU 0 adjusts PTE in MMU and restarts instruction. (2,2,3,3) is neither ≥ nor ≤ to (3,3,2,5)
•  CPU 1 writes to page 302 successfully. 2<3; 2<3; 3>2; 3<5

Page 43 Page 44

13. A linear compensating function adjusts a clock to: 14. A client gets a timestamp of 4:12:30.500 from a
a.  Tick at constantly decreasing or increasing time server. The elapsed time between the request
intervals until synchronization is achieved. and response was 20 msec (0.020 sec). The current
time on the client is 4:12:30.510. Using Cristian’s
b.  Compensate for variations in a clock’s frequency algorithm, what is the time set to on the client?
to make it appear to tick at a constant rate.
a.  4:12:30.480
c.  Tick at a constant faster or slower interval.
b.  4:12:30.490 Cristian’s algorithm sets the time
d.  Compensate for sudden changes such as leap
seconds or daylight savings time. c.  4:12:30.510 to:
d.  4:12:30.520 server time + ½(elapsed time)
= 3:12:30.500 + 0.020/2
= 3:12:30.500 + 0.010
= 3:12:30.510

Page 45 Page 46

15. Which mutual exclusion algorithm is useful when the 16. How does IP multicast achieve reliable packet
membership of the group is unknown? delivery?
a.  Centralized. a.  It doesn’t.
b.  Lamport’s. b.  Via best-effort multicast.
c.  Token ring. c.  Via atomic multicast.
d.  All of the above. d.  Via IGMP (Internet Group Management Protocol).
Lamport’s requires reliable group multicast.
Token ring requires the ability to construct a logical
ring of processes based on the group.
Centralized doesn’t require you to know about any
other process.

Page 47 Page 48

8
17. A copyset is used:
a.  In a DSM system that employs page caching.
b.  In a DSM system as an alternative to page
caching.
c.  To implement lazy release consistency.
d.  To implement entry consistency. The End.
A copyset is nothing more than a list of processors
that have a copy of a particular shared memory page.

Page 49 Page 50