Hardware Map
The linked file contains information about the hardware map. It gives the relation between hardware (DDL-Slot-ADC) and software numbering (Layer-Ladder-Module numbers).
SSD dead time
The following numbers apply only to the full configuration: 144 halfladders on 3 LDCs with 6, 5 and 5 DDLs each and a single GDC. The scaling to other configurations is not straightforward.
Full event | 668830 bytes |
Zero suppressed | 5278 bytes |
Occupancy 1% | 11914 bytes |
- Calculation of event size
- Each fired strip adds 4 bytes to the raw data stream
- The data size per module can be as large as 4x1536 = 6.144 KB/module (CM excluded - see below).
- The overall event size (CM excluded - see below) for the case of a 100% occupancy is 4x1536*1698 = 10.433MB.
- The CM offset is always sent in the raw stream in the current version of the firmware. Each CM channel adds 4 bytes to the stream and taking into consideration that we have 12 chips per module that makes in total 48 bytes/module.
- The overall size related to the CM channels is 4x12x1698 = 81.504 KB
- The final SSD event size is calculated as: Size(CM) + Size(Signal) = 10.514 MB
- In general the formula that gives back the event size is: Event size = Size(CM) + Nmodules x Nstrips x SizeOfStrip x Occupancy.
- Noise run: Our DA removes the 99.7% of the noise by imposing a 3σ threshold assuming that the noise is Gaussian. This means that statistically one would expect to find 0.3% of the strips fired even in a noise run. The expected dat size of such an event is:
- Size(CM) + 1698x1536x4x0.003/2 = 81.504 + 15.649 KB = 97.153 KB
- Occupancy 1%: Size(CM) + 1698x1536x4x0.1 = 81.504 + 104.325 KB = 185.829 KB
- Event size as a function of the ZS threshold
Trigger sequence | Dead time |
---|
L2a | 176 μs |
L2r | 101 μs |
L1r | 7 μs |
With these numbers one can calculate the maximum trigger rate (100 % busy) for any combination of triggers as long as the backpressure is not dominant.
With backpressure the notion of dead time becomes less useful, because of overlapping cycles in the various processes: front-end read-out, FEROM multi-event buffer, RORC buffers, LDC/GDC transfers. Moreover, in a multi-detector run the event building will be shared by all GDCs, regulated by the dead times of all detectors involved.
The basic data transfer rates are: DDL 200 MB/s, LDC – GDC 125 MB/s.
The following maximum trigger rate was measured in a stand-alone run with event building on, but without saving data either locally or on Castor: all suppressed 1.36 kHz L2a.
This compares to the GDC transfer limit of 1.48 kHz suppressed events that we can calculate from the numbers given above.
For full events the GDC transfers a maximum 12 Hz, which is compatible with the ~50 Hz we saw in the December configuration.