Advanced Programming in the UNIX® Environment

ptg10805159

786 ADatabase Library Chapter 20

Figure20.9 is a graph of the data from Figure20.8 for advisory fine-grained locking.

We plot the clock time as the number of processes goes from 1 to 16. We also plot the

user CPU time divided by the number of processes and the system CPU time divided

by the number of processes.

#processes

clock time

(seconds)

sys CPU/#proc,

user CPU/#proc

(seconds)

(^12345678910111213141516)
0
10
20
30
40
50
60
70
80
0
2
4
6
8

• ∗

Δ

• ∗

Δ

• ∗

Δ

∗

Δ

• ∗

Δ

• ∗

Δ

• ∗

Δ

• ∗

Δ

• ∗

Δ

• ∗

Δ

∗

Δ

∗

Δ

∗

Δ

∗

Δ

∗

Δ

∗

Δ

user CPU/#proc

system CPU/#proc

clock time

Figure 20.9 Values from Figure20.8 for advisory fine-grained locking

Note that both CPU times, divided by the number of processes, arelinear but that the

plot of the clock time is nonlinear.The probable reason is the added amount of CPU

time used by the operating system to switch between processes as the number of

processes increases. This operating system overhead would show up as an increased

clock time, but shouldn’t affect the CPU times of the individual processes.

The reason the user CPU time increases with the number of processes is that there

aremorerecords in the database. Each hash chain is getting longer, so it takes the

_db_find_and_lockfunction longer, on average, to find a record.

20.10 Summary

This chapter has taken a long look at the design and implementation of a database

library.Although we’ve kept the library small and simple for presentation purposes, it

contains the recordlocking required to allow concurrent access by multiple processes.

We’ve also looked at the performance of this library with various numbers of

processes using no locking, advisory locking (fine-grained and coarse-grained), and

mandatory locking. With a single process, we saw that advisory locking adds between