[
https://issues.apache.org/jira/browse/DERBY-4462?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12786401#action_12786401
]
Kristian Waagan commented on DERBY-4462:
----------------------------------------
Hi Bryan,
I know measuring elapsed time is not always as simple as one would like, and I
have some comments.
Regarding System.currentTimeMillis, I think the resolution ("update rate")
depends on the OS (and to some degree the hardware).
For instance, a rule of thumb for clock resolution used to be up to ~15 ms for
Windows, 10 ms for Linux 2.4 and 1 ms for Linux 2.6. I checked the resolution
on a Core 2 Quad system running Windows Vista, and it gave me 1 ms. So it seems
the situation is improving at least on some platforms.
Switching to System.nanoTime is a good move. One should realize that it doesn't
provide any guarantees that the resolution is better than for
System.currentTimeMillis, but in practice I believe the former provides a
higher resolution than the latter on modern systems.
Also, System.nanoTime is defined in relative terms, whereas
System.currentTimeMillis is defined in absolute terms. Values from the two
cannot be used together (i.e. System.currentTimeMillis() - (System.nanoTime()
/ 1000000) is not valid), and values from System.nanoTime cannot be used across
JVM boundaries.
> Use System.nanoTime on JDK 1.5 and higher to get finer precision internal
> measurements
> --------------------------------------------------------------------------------------
>
> Key: DERBY-4462
> URL: https://issues.apache.org/jira/browse/DERBY-4462
> Project: Derby
> Issue Type: Improvement
> Components: SQL
> Reporter: Bryan Pendleton
> Priority: Minor
>
> Derby uses System.currentTimeMillis() to compute the internal elapsed time of
> certain operations.
> For example, the parseTime, bindTime, optimizeTime, generateTime, and
> compileTime measurements
> that are computed by GenericStatement.prepMinion use this technique.
> System.currentTimeMillis is not terribly accurate, since it is only updated
> 60 times a second; this
> means that it doesn't give very good information about operations that take
> less than 16 milliseconds,
> and the compilation time of a SQL statement is often shorter than that.
> As Knut Anders observes, we could use System.nanoTime, which is available in
> JDK 1.5 and above,
> to make more precise measurements.
> See DERBY-4297 for some related observations.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
