CRU data format ALICE O 2 CRU 1

CRU data format ALICE O 2 -CRU 1

SPOILER ALERT CRU team would keep the word padding and these are the reasons … 2

DATA FORMAT (RAW MODE) GBT MODE 37. 5% extra padding 48 bit 80 bit WIDE MODE 12. 5% extra padding 16 bit 112 bit 3

DATA FORMAT (DET User Logic) 128 bit DET UL 0% extra padding 4

Experiment THROUGHPUT 5

Experiment THROUGHPUT TPC generates 90% of the total throughput in ALICE 6

DET CRU/FLP 7

DET User Logic CRU DET UL FLP UL = 430 (TOT 473) = 90 % no memory loss = 215 (TOT 247) = 87 % no memory loss 8

TPC RAW MODE GBT WORD 112 bit Throughput 112 bit @ 40 MHz = 4. 48 Gb/s 20 GBT link x CRU = 89. 6 Gb/s 100 us @ 10 Hz = 90 Mb/s Extra padding GBT WORD 128 bit Throughput 128 bit @ 40 MHz = 5. 12 Gb/s 20 GBT link x CRU = 102. 4 Gb/s 100 us @ 10 Hz = 100 Mb/s 2 CRU x FLP = 200 Mb/s 3 CRU x FLP = 300 Mb/s 9

CONCLUSIONS • During PHYSICS data taking the UL is used, so no extra padding in 90% of the CRU • When the RAW mode is used the extra padding is added, however in this case we run in trigger or throttling mode. We lose the 12. 5 % of memory for limited amount of data (requirement for DEBUG/CALIBRATION mode is 100 Mb/s) • DETECTOR that will have always the padding in place are ITS, MFT, TOF. Their contribution in terms of data throughput is < 60 GB/s. • The work to change the data format is not trivial: • consequence in the firmware, more complexity and this is for 3 different type of usage wide continuous, GBT continuous and GBT packets (3 different packers and size computer), • consequence in the software, data are not easily readable anymore, even for CRU tests, • the modification will affect all the CRUs but only a 10% will actually benefit from it (the ones that don’t have DET UL). http: //svnweb. cern. ch/world/wsvn/alicetdrrun 3/Notes/Run 34 System. Note/detector-readalice/ALICErun 34_readout. pdf 10

QUESTIONS • How the extra padding affect the reconstruction in terms of computing power? • The TPC will have local processing in the FLP anyway. If the extra padding is an issue when running in DEBUG mode, it could be removed by a software before shipping the data over the network. • What is the amount of DATA that the experiment will collect in DEBUG mode over the year compared to PHYSICS DATA? 11