Experimental Physics and
Industrial Control System

Steve Lewis <[email protected]> · Sat, 16 Nov 2002 13:26:43 -0800

The Channel Watcher moves the save part of save/restore from the IOC
to UNIX.

This an old subject.
The motivation for the "bumpless" IOC-based approach to save/restore is:
 -- reduce to a minimum the upset caused by rebooting an IOC while it
is being used in an operational facility
 -- maintain strong EPICS architectural bias against single-point
failures and central bottlenecks
 -- minimize administrative work of synchronizing (multiple,
differently formatted) lists of things about channels
Bumpless assumes:
 -- an application suite such as BURT is used to manage grand
collections of set-points for initiating and changing how the facility
operates, either from a non-operating state or when shifting over
 -- you have got NFS working well for the usual booting-up process on
your IOCs.
Bumpless provides:
 -- a very simple way to mark fields in your database as "bumpless";
just another attribute like PP or MDEL to be naturally managed by the
knowledgeable application developer
 -- it has a few parameters to trade off efficiency with network/IOC
loading: how often to save/now much to "batch" saves
 -- since the single, short file it uses is typically kept on the same
file system where the database loads from, the success of the one more
or less ensures the success of the other
 -- VxWorks NFS uses UDP; all told there is not much overhead. (CA,
which is very efficient in most respects, uses TCP; and Channel Watcher
may be additionally using NFS to access the file system from whichever
host it is actually runnning on)
 -- it is a very small amount of code; it is easily configured in to
the IOC initialization; it does not require management of yet another
central client service (get/build right code version; make sure it
starts/restarts on some server; make sure its to-do list is updated
when *ANY* IOC database is changed)

Disclaimer: I had a lot of involvement with getting the bumpless
capability going (although Bob and Carl and a few others did all the
work); and I am working with a very large non-EPICS control system in
which almost everything depends on central servers/services.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System