Experimental Physics and
Industrial Control System

Maren Purves <[email protected]> · Mon, 17 Sep 2007 12:31:09 -1000

Emmanuel Mayssat wrote:

John,
There are several issues with power cycling all of which I experienced
at one point.
1/ The operator needs to know which unit to power cycle. Given the
distributed nature of EPICS, this may not be obvious.

that's why it's a good idea to give iocs and records and everything
else involved names that mean something.
2/ During power cycle procedure you do not have any control, so the
synchrotron is in an unknown, uncontrollable state which is exactly what
I want to avoid.

but, if it's locked up you are already in that state.
3/ A power cycle resets all output signals (to zero) even if they are

later restored to their previous values. This is particularly true for

DAC PCI cards with AO and BO.

If you don't want to power cycle, there's such things as console
switches (I guess that's what you're talking about below) as well
from which you can do (and watch) soft reboots.
A console server would be a great complement as well. But PDU and

console servers do not guarantee I know the state of my synchrotron at

all time.

In real life you never can. The best you can do is make sure that when
it fails it fails in a safe manner. Hardware breaks by itself, and power
outages aren't always avoidable (or predictable) either. (And
earthquakes?)

Maren
On Sat, 2007-09-15 at 09:08 -0400, John Sinclair wrote:

Emmanuel,
I, for one, never rely on an EPICS IOC to protect property or personnel.
If an IOC

experiences a software-related lock-up, I cycle its power. If it resides
in an

inaccessible location, I connect to a special network-attached device
via telnet

to perform the power cycle. In the worst case, this causes a small change in

some output (values are saved every ten seconds so we can loose a change

that occurs immediately before the failure).
John Sinclair

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System