User manual
Appendix
ComTec GmbH 7-12
(whether from an ADC or any other front end device that produces digital data), both with and without the
PC graphic real-time display turned on (i.e. writing data only to hard disk, and writing to hard disk plus
histogramming in RAM for a real time display)
A: no data were written to hard disk, only transfered into the PC RAM.. Histogramming and writing to hard
disk depends on the performance capabilities of the PC used.
Q: But now I am wondering if my understanding is correct, since at the bottom of the chart states "List
Data to SCSI (hard disk I assume)" and "List data to IDE". What is the difference? These numbers are
MUCH lower 1600K and 120K?
A: Yes this was the observed throughput limit when writing a listfile to a standard harddisk with IDE or
SCSI interface. Of course these numbers can be improved better by using a faster (and naturally more
expensive) hard disk.
Q: Isn't the other data (for IK, 4K, 16K FIFO's etc) for writing to a hard disk? (of some type). Nothing really
says. If not, then what is it for?
A: No, it is for histogramming spectra in the PC RAM.
Q: I hope this main data is not only for using a RAID disk or something exotic like that.
A: We had no RAID system, but if you want to transfer data to disk with the speed that is in principle
possible, you need a RAID system or anything else that can accept such a data rate..
Q: C. Why does the dual processor PC make such a difference? I would think the PCI board would be the
limitation. Pls explain.
A: The evaluation of the data for histogramming needs some processor time and therefore it helps very
much to have a dual processor system, as one processor in principle can handle the transport of the input
data and the other evaluate the data.
Q: D. Do the new "100 MHz bus" PC's make any difference in the throughput? Are they an advantage?
How?
A: Yes. But we used a 100 MHz bus PC for this measurement.
Q: E. Is this only for periodic-arrival statistics data? versus random arrival?
A: This was for periodic data, as no random data was available - but since we buffer the data there
should be no difference.
Q: 1. Thank you for the clarification on the performance chart. That was not at all clear. I think the chart
also needs to say that this is RAM histogram data explicitly and identify the type of PC used (speed, etc).
A: 1) the performance chart is actually quite accurate. If RAM histogramming, a RAID system or any
other storage device used that is fast enough to accept this data is immaterial. Every few months there
are new, faster storage devices being introduced which is why we do not intend to make statements that
puts qualifications on our performance that will be outdated a few weeks later. It is just a question of what
the customer will spend for his storage device that determines the sustainable data storage rate, that his
system has. In essence it is not the MPA-3 system that limits the data transfer rate but the storage device
the customer is willing to buy.
Q: 2. I need to know if the "List Data to..." performance numbers at the bottom of this chart also improve
by a 4X if only 4 parameters are used??.
A: 2) The transfer rate we listed is the maximum rate that is possible without loosing data. The transfer is
not influenced by the number of ADCs connected. It can be one or 16 or anything inbetween. If you have
just one ADC and the max. transfer rate to your storage device is for example 400,000/s it is quite clear. If
you have two ADCs then the total conversions from both ADCs can not exceed 400,000/s - don't think of
any restrictions like 200,000/s per ADC - it does absolutely not matter what the actual counting rate of
each ADC is as long as the sum does not exceed 400,000/s. In coincidence operation the conversions
that meet the coincidence conditions can not exceed 400,000/s even if the input rate to the ADCs is
perhaps much higher. As you know normally not all input events produce a coincidence!!
Q: 3. It is not clear to me why the system performance improves so much, with periodic data, as the FIFO
size increases. For burst data, I could understand that, but for periodic data, I do not understand. It must