[15 poitns] Consider two tables, R(A,B) and
S(C,D), where C is a key in S, and the selection-join operation sA=v(R)
⋈B=C
S. Assume that a record is
a large as a block and that
T(R)=B(R)=B(S)=10000. There is a clustered index on S.C, and we
need four disk accesses to lookup a
value in that index (including the disk access to get the record). We consider the following methods for
the join:
- block nested loop join
- partitioned hash-join
- index join
Compute the I/O cost for each join method in each of the
four cases below. Indicate in each of the four cases the best join
method.
|
V(R,A) = 1 |
V(R,A) = 10000 |
M = 101 |
|
|
M = 5002 |
|
|
[15
points] After a system’s crash, the undo-log using non-quiescent checkpointing
contains the following data:
a. [5
points] What are the correct values of the three <START CKPT ????> records ? You have to provide three correct
values for the three “????”s.
b. [5
points] Assuming that the three <START CKPT ???> records are correctly stored in the
log, according to your answer at a., show which elements are recovered by the
undo recovery manager and compute their values after recovery.
c. [5
points] Indicate what fragment of
the log the recovery manager needs to read.