Databases Reference
In-Depth Information
Table 1: The number of nodes in a Btree for various numbers of streams
Similarly to the insert operation, the response time of a delete operation can be
reduced by asynchronous backup:
(14.4)
Retrieve:
Because a retrieve operation does not require logs and backups, the cost
of taking log and backup should be eliminated. Therefore, the cost formula becomes:
(14.5)
Of course, by assuming some cache mechanism, those costs will be reduced further.
For instance, disk accesses during index traversal can be omitted by using a cache.
In that case the response times for insert, delete and retrieval operations become
as follows:
(14.6)
(14.7)
(14.8)
3.2.2
The Cost of Operations in Autonomous Disks
First, we derive the probability that a disk contains the target node, to estimate the
frequency of communication between disks. If we adopt a Fat-btree as the
distributed directory, the probability that a PE has a node in level
h
that is necessary
for the traverse is:
Here,
N
n
(h)
is the total number of nodes for level
h
in a Btree. Table 1 lists values
for a variety of stream numbers
N
s
.
11)
