C ARNEGIE M ELLON U NIVERSITY

D EPARTMENT OF C OMPUTER S CIENCE

15-445/645 – DATABASE S YSTEMS (FALL 2017)
P ROF. A NDY PAVLO

Homework 6 (by Sivaprasad Sudhir) – Solutions

Due: Monday Nov 27, 2017 @ 11:59pm

IMPORTANT:
• Upload this PDF with your answers to Gradescope by 11:59pm on Monday Nov 27,
2017.
• Plagiarism: Homework may be discussed with other students, but all homework is to be
completed individually.
• You have to use this PDF for all of your answers.
For your information:
• Graded out of 100 points; 3 questions total
• Rough time estimate: ≈ 1 - 2 hours
Revision : 2017/12/10 13:52

Question Points Score

Write-Ahead Logging 35
ARIES 37
Miscellaneous 28
Total: 100

1
15-445/645 (Fall 2017) Homework 6 Page 2 of 7

Question 1: Write-Ahead Logging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . [35 points]

Consider a DBMS using write-ahead logging with physical log records with the STEAL and
NO-FORCE buffer pool management policy. Assume the DBMS executes a non-fuzzy check-
point where all dirty pages are written to disk.
Its transaction recovery log contains log records of the following form:
<txnId, objectId, beforeValue, afterValue>
The log also contains checkpoint, transaction begin, and transaction commit records.
The database contains three objects (i.e., X, Y, and Z).
The DBMS sees records as in Figure 1 in the WAL on disk after a crash.

LSN WAL Record

1 <T1 BEGIN>
2 <T1, X, 1, 2>
3 <T2 BEGIN>
4 <T3 BEGIN>
5 <T2, Y, 1, 2>
6 <T2 COMMIT>
7 <T1, Y, 2, 3>
8 <T3, Z, 1, 2>
9 <CHECKPOINT>
10 <T1, X, 2, 3>
11 <T1, Y, 3, 4>
12 <T3, Z, 2, 3>
13 <T3 COMMIT>
14 <T1, Z, 3, 4>

Figure 1: WAL

(a) [10 points] What are the values of X, Y, and Z in the database stored on disk before the
DBMS recovers the state of the database?
2 X=1, Y=1, Z=1
2 X=1, Y=2, Z=3
2 X=2, Y=3, Z=2
2 X=2, Y=2, Z=3
2 X=3, Y=1, Z=2
2 X=3, Y=4, Z=3
2 X=3, Y=4, Z=4
 Not possible to determine
Solution: The checkpoint flushed everything to disk, but then all the data objects were
modified by transactions after the checkpoint.
Since we are using NO-FORCE, any dirty page could be written to disk, so therefore we
don’t know the contents of the database on disk at the crash.

Question 1 continues. . .
15-445/645 (Fall 2017) Homework 6 Page 3 of 7

(b) [5 points] What should be the correct action on T1 when recovering the database from
WAL?
 undo all of T1’s changes
2 redo all of T1’s changes
2 do nothing to T1
Solution: T1 never committed. All of its changes should only be undone.

(c) [5 points] What should be the correct action on T2 when recovering the database from
WAL?
2 undo all of T2’s changes
2 redo all of T2’s changes
 do nothing to T2
Solution: T2 committed before the checkpont. All of its changes were written to disk.
There is nothing to redo or undo.

(d) [5 points] What should be the correct action on T3 when recovering the database from
WAL?
2 undo all of T3’s changes
 redo all of T3’s changes
2 do nothing to T3
Solution: T3 committed after the checkpont, so that means the DBMS has to redo all
of its changes.

(e) [10 points] Assume that the DBMS flushes all dirty pages when the recovery process
finishes. What are the values of X, Y, and Z after the DBMS recovers the state of the
database from the WAL in Figure 1?
2 X=1, Y=1, Z=1
 X=1, Y=2, Z=3
2 X=2, Y=3, Z=2
2 X=2, Y=2, Z=3
2 X=3, Y=1, Z=2
2 X=3, Y=4, Z=3
2 X=3, Y=4, Z=4
2 Not possible to determine
Solution: X = 1 (rollback the beforeValue from T1)
Y = 2 (rollback to the afterValue made by T2)

Question 1 continues. . .
15-445/645 (Fall 2017) Homework 6 Page 4 of 7

Z = 3 (rollback to the afterValue made by T3)

Homework 6 continues. . .
15-445/645 (Fall 2017) Homework 6 Page 5 of 7

Question 2: ARIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . [37 points]

Consider a DBMS using ARIES protocol for logging records.
Its transaction recovery log contains log records of the following form:
<txnId, objectId, beforeValue, afterValue>
The log also contains checkpoint begin, checkpoint end, transaction begin, transaction commit,
transaction end and undo action records.
The database contains four objects (i.e., X, Y, Z and W).
The DBMS sees records as in Figure 2 in the log on disk after a crash. Notice that there are
no dirty pages nor active transactions, during the first and only checkpoint.

LSN ARIES Record

1 <BEGIN CHECKPOINT>
2 <END CHECKPOINT>
3 <T1, X, 1, 2>
4 <T2, Y, 3, 4>
5 <T1 COMMIT>
6 <T1 END>
7 <T3, Z, 5, 6>
8 <T4, W, 5, 6>
9 <T2 COMMIT>

Figure 2: ARIES

(a) [10 points] Which transaction(s) will be undone, if any?

2 T1
2 T2
 T3
 T4
2 None of the above
Solution: T3, T4

(b) After the recovery has ended successfully and assuming that all the dirty pages have been
flushed to disk,
i. [3 points] What will be the value of X on disk?
2 1
 2
2 Unknown
Solution: New value of X will be on disk as T1 will be redone

ii. [3 points] What will be the value of Y on disk?

Question 2 continues. . .
15-445/645 (Fall 2017) Homework 6 Page 6 of 7

2 3
 4
2 Unknown
Solution: New value of Y will be on disk as T2 will be redone

iii. [3 points] What will be the value of Z on disk?

 5
2 6
2 Unknown
Solution: Old value of Z will be on disk as T3 will be undone

iv. [3 points] What will be the value of W on disk?

 5
2 6
2 Unknown
Solution: Old value of W will be on disk as T4 will be undone

(c) [15 points] After the recovery has ended, what will be the contents of the log? Specify
the log records after the crash.
Solution: 10: <T2 end>
11: <CLR undo T4 LSN 8>
12: <T4 end ;>
13: <CLR undo T3 LSN 7>
14: <T3 end>

Homework 6 continues. . .
15-445/645 (Fall 2017) Homework 6 Page 7 of 7

Question 3: Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . [28 points]

All the questions below refer to recovery algorithms that use a write-ahead log (WAL). State
True or False for each of the following questions.
(a) [4 points] Under NO-STEAL + FORCE policy, a DBMS will never need to redo the
changes of a committed transaction during recovery.
 True
2 False

(b) [4 points] Under NO-STEAL + FORCE policy, a DBMS will have to undo the changes
of an aborted transaction during recovery.
2 True
 False

(c) [4 points] Under the NO-STEAL policy, a DBMS will need to store the whole table in
RAM if a transaction updates all the records of that table.
 True
2 False

(d) [4 points] While doing deferred updates with WAL, if we prevent the DBMS from writ-
ing dirty records to disk until the transaction commits, then we do not need to store their
original values.
 True
2 False

(e) [4 points] Most production systems use STEAL + NO-FORCE policy as it has the
fastest recovery performance.
2 True
 False

(f) [4 points] In ARIES, log records are immediately flushed on the log, as soon as they are
produced.
2 True
 False

(g) [4 points] Without checkpoints, the redo phase of ARIES recovery should process the
whole log.
 True
2 False

End of Homework 6