blob: 6696a50c5655c2a16ca0ae74859daa1f4516ee9b (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
|
* Opgave 8.3 fra bogen
#+begin_quote
Consider the parallel hash join algorithm in Sect. 8.4.1.2.
Explain what the build phase and probe phase are.
Is the algorithm symmetric with respect to its input relations?
#+end_quote
** Explain build and probe phases
The build phase hashes the =R= phase into $p$ partitions.
Then it sends the partitions to their respective nodes.
The nodes in $[1, p]$ each receive the partitions and creates a local hash table for $R_j$.
In the probe phase all nodes do the same with =S= and
the nodes $[1, p]$ receives the partitions of =S= joins it with =R=.
** Is the algorithm symmetric?
No it's not, be inner join(thus =R=) in the build phase, must be completely stored and hashed.
It can't start doing join on it while it's receiving.
A symmetric algorithms allows changing the order in which inputs are consumed.
* Opgave 8.7 fra bogen
#+begin_quote
Consider a nine way join (ten relations are to be joined),
calculate the number of possible right-deep, left-deep, and bushy trees,
assuming that each relation can be joined with anyone else.
What do you conclude about parallel optimization?
#+end_quote
With right-deep we can join them in $10!$ different ways, as we just order them differently down the spine.
The same for left-deep.
For bushy trees we look at the leafs of the tree.
Here we can have $10!$ different orders again.
But we multiply this with the number of ways to create a 5 leaf binary tree, which i count as 3.
Therefore the end result is $2 * 10! + 3 * 10! = 5 * 10! = 18144000$.
Hmm that does not seem right.
|
