Experimental Physics and
Industrial Control System

[email protected] (Bob Dalesio) · Tue, 13 Jun 95 08:42:25 MDT

There has been some discussion about the possibilty of producing a
lite database. This need arises when high rep rates are required.

Currently, the database record processing averages around 85 useconds
per record. This includes input (if any), conversions (if any),
alarm checking, monitors and forward links.

Previously, we discovered that the processing of the data links had
some room for improvement. Matt Needes provided the modification
for "fast" data links. This is a step in the right direction, but
still leaves record processing out of the KHz scanning range.

As a test, it has been proposed at several sites that a "lite"
database record be produced. The overhead that would be
taken out is:
conversions to floating point,
all limits would be in integer 
a rate would be specified for alarm checking and posting monitors 
(optionally - no monitors, no alarm checking)
alarm and monitor routines would be - inlined
only the I/O subsystem would be called directly - assuming VME, the
	device layer should include the VME interface (i.e. no driver)
If anyone has ideas for removing more overhead - please post it.

We have no time to run this experiment, however. If anyone is
undertaking this, please post your progress to tech-talk. There
are more than a couple of applications that could use the KHz scanning
range. It may be that the scheduling overhead for the database is too
high. In any case, it would be good to find out the maximum rate we
can scan a minimal database record.

	.....Bob

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System