aboutsummaryrefslogtreecommitdiff
path: root/sem7/db/lec5.org
diff options
context:
space:
mode:
Diffstat (limited to 'sem7/db/lec5.org')
-rw-r--r--sem7/db/lec5.org42
1 files changed, 42 insertions, 0 deletions
diff --git a/sem7/db/lec5.org b/sem7/db/lec5.org
new file mode 100644
index 0000000..6696a50
--- /dev/null
+++ b/sem7/db/lec5.org
@@ -0,0 +1,42 @@
+* 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.