Databases Reference
In-Depth Information
Figure 3: Packet level retrieval termination detection
2, unknown > and < 3,2, unknown > in T, when < unknown, 2, unknown > is
received, these three branch histories will also be merged into < 2, unknown >.
The changes to T are shown in the right-hand side of Fig.3. Note that merging is
impossible even if c is received and is postponed until d having the fixed first term
of the branch history will be received.
Lastly, we estimate the overhead of communication cost due to attaching branch
histories. Let n be the number of the objects retrieved in a PE, O S be the size of an
OID, H S be the size of a branch history, and P S be the size of a packet. Whereas
object-level termination detection requires a branch history for each object, packet-
level detection requires a branch history for each packet. Let ß, ß o , and ß p be the
number of packets required in no termination detection, object-level detection,
and packet-level detection, respectively. For example, let O S be 8 bytes, P S be 256
bytes, and H S be 12 bytes. The required numbers of packets can be calculated as
shown in Table 1, which indicates that the communication overhead caused by
the packet level detection is very small.
Table 1: Number of required packets
3
A database programming language based on the object-
parallel approach
Parallelism can be exploited not only in the retrieval phase of the database man
agement system (DBMS), but also in the application processing phase after the
retrieval. If we can develop a parallel object-oriented programming language for
Search WWH ::




Custom Search