sPHENIX EMCal discussion

[pinkenburg <pinkenburg AT bnl.gov>]

    Hello Jamie, Bill
    
    so far we have been working with 300kB/event for "non tpc" data:
    
https://indico.bnl.gov/event/15770/contributions/63203/attachments/40979/68555/Soltz_computing_plan.pdf
    
    From the offline point of view, if this changes within reason (not
    by an order of magnitude), it'll just affect how much tape we need
    for storage of the raw data. Our available bandwidth and disk i/o
    should be able to handle increases easily. Technically this would
    really just mean fewer events in the 20GB calorimeter data files so
    we just need to read more bytes into the calorimeter reconstruction
    for the same number of events.
    
    Chris

On 6/23/2022 1:50 PM, Jamie Nagle wrote:

Hello Bill (cc EmCal list and Chi),

A couple of years ago, John Haggerty and I did a set of studies on DAQ throughput for the calorimeters.    With 0-20% central HIJING events run through GEANT, the EMCal average tower energy was 37 MeV and with a threshold for zero suppression of 30 MeV, there was a 28% occupancy.     

At the time, the advice was to have 3 ADC boards/XMIT for the EMCal and 2 ADC boards/XMIT for the Hadronic Calorimeter, and then 1 DCM2 / PAR III --> jSEB2d board.    That would enable one to run the EMCal with 16 samples and "light" zero suppression and the Hadronic Calorimeter without zero suppression.   Those numbers did not translate into additional orders.

The reality is that at best we will have 3 ADC boards / XMIT for both the EMCal and the HCal and 2 DCM2 / PAR III -> jSEB2d board.    Thus, the bandwidth limitations are exactly the same for the EMCal and HCal -- except considering the total final data volume going to RCF.

I am glad to re-show my earlier presentation, but the summary is that there is a Dual Port memory on the DCM II that when it gets 75% full, the busy is raised.    The main bandwidth limitation is going to be the 120 MHz x 36 bits on the DCM II token passing to the PAR III.     I have a "leaky pipe" simulation that includes occupancy, centrality fluctuations, and stochastic modeling for the multi-event buffering etc.    

(1) If you run with 16 samples and try to push 15 kHz Level-1 trigger rate through the DAQ, you get the following without zero suppression.

Thus you have 33% livetime -- i.e., you keep pushing too hard and the buffers keep filling up, and you get a 5 kHz rate.    I would not call this "grinding to a halt."

(2) If you run with 16 samples and try to push 15 kHz Level-1 trigger rate through the DAQ with 60% occupancy in central events (overestimating) and scale occupancy with Nch one gets.

Also, the zero suppression in the DCM II will be something like one sample minus another sample (over an average of a couple of pre-samples).    There is no bandwidth in the DCM II to do a sophisticated pulse shape / fit.    The algorithm Chi is working on will keep two samples (a pre and on the peak) even if the channel is zero suppressed, and keep all N = 16 (?) samples if it is above the zero suppression.

Note that there are a lot of real world problems not included here, but this should capture the basic bandwidth landscape once things are fully working (which may take most of the run).

Chris may want to comment on the implications on the total data volume.    

Sincerely,

Jamie

||------------------------------------------------------------------------------------------
|| James L. Nagle   
|| Professor of Physics, University of Colorado Boulder

|| EMAIL:   jamie.nagle AT colorado.edu
|| SKYPE:  jamie-nagle        
|| WEB:      http://spot.colorado.edu/~naglej 
||------------------------------------------------------------------------------------------

On Thu, Jun 23, 2022 at 10:47 AM W.A. Zajc <zajc AT nevis.columbia.edu> wrote:

Hello all:

I have conflicts that prevent attendance to the EmCal meetings, so call me an interested non-observer.

I have a question about the EMCal plans for zero suppression in Year-1. This was sparked by a statement from Chris Pinkenburg at Tuesday’s Calibrations meeting that always reading out non-suppressed calorimeters will slow the readout to a crawl (Chris, please correct if I am misquoting you). 

A back-of-the-iPad calculation shows that in central Au+Au, the average energy in a delta-eta x delta-phi = 0.025 x 0.025 is about 50 MeV. Divide this by ~4 for MB. These numbers immediately raise the question of what energy threshold is being considered for zero suppression, and how do various physics signals depend on this threshold?

1) Yes, I know the threshold will be on an ADC value, not an energy. But I think its still useful to phrase the question in terms of truth values.

2) Yes, I know that a central collision does not uniformly illuminate the EmCal with 50 MeV hits; it will be much more granular and concentrated with the average energy per *hit* tower ~ 0.3-0.4 (??) GeV. 

Chi has reminded me that when zero suppressing you are not just throwing away empty words but you are also adding bits to specify the tower ID. So real zero suppression gains have to be determined with non-trivial simulations. Perhaps this has already been done for the EMCal?

Best regards,

Bill 

—————————————————

W.A. Zajc I.I. Rabi Professor of Physics Columbia University New York, NY 10027

https://blogs.cuit.columbia.edu/waz1/ —————————————————

On Jun 23, 2022, at 8:22 AM, Craig Woody <woody AT bnl.gov> wrote:

Dear All,
  Just a reminder that we will have our regular EMCAL meeting tomorrow, Friday June 24th starting at 9:00 am EDT. Here's the Indico link: https://indico.bnl.gov/event/16262/

Cheers,
Craig
_______________________________________________
sPHENIX-EMCal-l mailing list
sPHENIX-EMCal-l AT lists.bnl.gov
https://lists.bnl.gov/mailman/listinfo/sphenix-emcal-l

_______________________________________________
sPHENIX-EMCal-l mailing list
sPHENIX-EMCal-l AT lists.bnl.gov
https://lists.bnl.gov/mailman/listinfo/sphenix-emcal-l
    
    
-- 
*************************************************************

Christopher H. Pinkenburg	;    pinkenburg AT bnl.gov
				;    http://www.phenix.bnl.gov/~pinkenbu

Brookhaven National Laboratory	;    phone: (631) 344-5692
Physics Department Bldg 510 C	;    fax:   (631) 344-3253
Upton, NY 11973-5000

*************************************************************