Sheila, Jonathan, Dave:
Sheila had a question about the relative timing of an analog event which was recorded by both a front end computer and a Beckhoff slow controls computer. When looking at the two channels from the DAQ, there was a time difference of about a tenth of a second.
As a test, I have some slow signals coming from front end models and being a acquired by the DAQ through two paths: one direct alongside the fast data, and an equivalent channel coming in via the EDCU. To show the delay, I looked at the h1ioplsc0 model's state word, and manually changed it by running an excitation to set the EXC bit. In the attached plot, the signal being sent directly to the DAQ data-concentration is the upper red plot, the signal being acquired over the network via the EDCU is the lower purple plot. As can be seen, the front end data is packetized into a 1/16th block before sending to DC, which in turn packetizes the data into a 1/16 block and adds the Beckhoff data point. This results in an apparent 2 sample delay (0.125S) in the red plot. In actuality, the red plot is timed correctly, and the Beckhoff event is incorrectly time stamped as preceding the event by 0.1 S.
We think moving the EDCU to a front end would resolve this.
The use of a front-end based EDCU (instead of the DAQ Data Concentrator) is already implemented in the CDS code base. It has already been tested extensively on test stand and, and briefly on L1. There are installation instructions in the DCC. This change is required to ensure EPICS data agreement when multiple Data Concentrators are used to increase redundancy.
Subsequent inquiries concerning my alog has made me realize my entry could be made more clear. The issue is that slow data coming directly from the front end computer to the DAQ data concentrator has a pipe-line delay, whereas data being acquired by the EDCU does not. When looking at the two signals in real time mode, the EDCU signal changes first, followed 3 cycles later (0.18S) by the FEC channel. The data-concentrator takes the pipe-line delay into account by timing the signal from the FEC's time stamp (not the EPICS IOC), meaning that when this event is later replayed by opening frame files, the FEC signal has the correct time, and the EDCU signal erroneously appears to have happened before the event time.