Database Reference
In-Depth Information
The following measurements were made for just one “gray” example. In an
installation with 100 mips processors, a SELECT had this profile when the index
used wasn't even semifat.
TR
=
10
,
000
TS
=
10
,
000
F
=
1000
RS
=
1000
275 ms
A semifat index reduced TR to 1000, the eliminated random touches being
skip-sequential, while TS, F, and RS remained the same. The CPU time was
reduced by over half to 124 ms; the CPU time per skip-sequential read in this
case would seem to be (275
CPU time
=
−
124 ms)/9000
=
17
s, illustrating the benefits
µ
that skip-sequential provides.
CPU Time Still Matters
Reducing the number of touches—all kinds of touches, not only the heavy ran-
dom touches—can bring CPU time down significantly. While the random read
time is normally the dominating component of the response time, saving CPU
time is still important in many installations. Mainframe CPU time may cost
U.S.$1000 per hour—largely because the monthly charge for many software
licences depends on the installed mips (CPU power). With LINUX, UNIX, and
Windows platforms, the cost of CPU time is much lower as we have seen, but
saving CPU time may be important to avoid hitting a server's maximum pro-
cessor capacity. Splitting the load across several servers always introduces CPU
and storage overheads. It also increases the risk of lock waits.
Search WWH ::
Custom Search