scalable tridas for the nemo projectricap09.roma2.infn.it/slides/chiarusi_ricap2009r.pdf · n. of...

Scalable TriDAS for the NEMO Project

Tommaso ChiarusiINFN Sez. Bologna & Phys. Dep. Univ. Bologna

for the NEMO Collaboration

Monte Porzio Catone, Villa Mondragone 14/05/2009

Talk overview

Neutrino Telescope: just add water...all data to shore: challenging throughput TriDAS base conceptFrom NEMO Ph.1 to the scalable km3 TriDAS architecture

Tommaso Chiarusi RICAP 2009

Cherenkov light

! "

Upgoing "

seabed depth > 3000 m

Atmospheric downgoing "

Less atm. bkg. when deeper ( below 3000 m usl)Expected signal > atm. bkg. when E!>10 TeV !Telescope Vol > 1 km3 for some evts/year ! Many PMTs (> 5000) ! complex experimental setup!NEMO location: CAPO PASSERO @ 3400 m usl

See Paolo Piatteli’s talk in plenary session todayNEMO

Capo Passero Site

100 km


some km3 detector possible layouts:

Det.Unit

cable PJB-SJB

cable SJB-DU

Square Grid

Hexagonal Grid90 Detection Units80 PMTs / D.U.7200 PMTs

81 Detection Units80 PMTs / D.U.6480 PMTs total


One muon event (only muon hits shown, no background)theTelescope is made of

DETECTION UNITS

20 floors, with 4 10” PMTs each


Expected Optical Data sources bioluminescence (neglig. @ depth !2500 m) upgoing neutrinos (atm. + signal) (50 day-1 / km3) atmospheric muons (10÷100 Hz / km3) 40K decays (30÷40 kHz / 10” PMT @ 0.5 p.e )

Hit samples Vs. pulse time on a 10” PMT with 0.5 p.e. thr. (NEMO Ph.1 data)

S.P.E. wave form(~16 samples = 16 Bytes)

S.P.E. HIT SIZE:Hit PMT Info

+ Hit Time

+Hit Charge

+Hit Wave Form(samples)

28 BytesTommaso Chiarusi RICAP 2009

Expected Data Rates10” PMT

single rate(kHz)

Data Rateper PMT(Mbps)

Data Rate per Floor*

(Mbps)

Data Rateper D.U.**

(Gbps)

Data Rate per km3 ***

(Gbps)

40 8.8 35.2 0.7 70

80 16.8 67.2 1.3 130

150 32 128 2.5 250

300 64 256 5.0 500

** 20 Floor/D.U. * 4 PMT/ Floor ** 100 D.U.


(bare 40K)

(NEMO Ph. 1)

(present DAQ)

(expanded DAQ )

The TriDAS reduces the data rate by filtering the data stream bunched in Time Slice (TS).

A TS contains all data from all ( or part of) the detector occurred in a given time interval (~100 ÷200 ms).

WHAT IS SCALABLE?

The TriDAS principal elements:

- Trigger System Controller (TSC): monitors and serves the TriDAS - (1 GbE I/O)

- Hit Managers (HM): receive optical data; prepare and distribute the TS to the TCPU

- TriggerCPUs (TCPU): apply the trigger logics to the assigned TS and transfer the selected data to the EM;

- Event Manager (EM): receives the triggered data from TCPUs and build the Post-Trigger events data file

Atmospheric Muon Signal (@ 3000 m depth) assuming! - Rate µ atm:! ! 100 Hz

! ! ! - RateK40 : ! ! 300 kHz

! ! ! - N. PMTs:! ! 8000

! - Rec. time window:! 6 µs

Post-Trigger data rate: ~ 40 MBps

Stored data /day ~1 TB !!!!

Upgoing Neutrino Signal assuming! - Rate ! :! ! < 4 10-3 Hz

! ! ! - RateK40 : ! ! 300 kHz

! ! ! - N. PMTs:! ! 8000

! ! - Rec. time window:! 6 µs

Post-Trigger data rate: ! 2 kBps

Stored data /day ! 150 MB !!!!


HM 1 HM 1

HM 1 HM 1

HM 2 HM 2

HM 2 HM 2

HM 3 HM 3

HM 3 HM 3

TCPU 1 TCPU 1

TCPU 1 TCPU 1

TCPU 2 TCPU 2

TCPU 2 TCPU 2

TCPU 3 TCPU 3

TCPU 3 TCPU 3

DU-TS i, 1 DU-TS i, 2 DU-TS i, 3

Full TS i

Time slice and Barrel Shift Paradigm

If a TCPU needs more time, just add one more!

[refer to: M.F. Letheren, 1995 CERN School of Computing]


TriDAS for NEMO Ph.1

Gbit switch

Master CPU <70 MB/s

Hard

Disk

Hard

Disk

CPU

PC Floor 1

PC Floor 2

< 70 MB/s

Windows

Linux

PC Monitor

DM - RC

PC Floor 3

PC Floor 4

The maximum throughput of the MiniTower was " 512 Mbps

One Machine, the Master CPU, played all the rules: Hit Manager, Trigger, Event Manager

CABLE

Floor 1

Floor 2

Floor 3

Floor 4


TriDAS for a first prototype Detection Unit with 16 Floors; expected through put " 2 Gbps standard 1 GbEthernet Networking

Floor 1

Floor 2

Floor 3

Floor 4

Floor 5

Floor 6

Floor 7

Floor 8

Floor 9

Floor 10

Floor 11

Floor 12

Floor 13

Floor 14

Floor 15

Floor 16

Shore

interfaces to:


Possible Network infrastructure for km3 TriDAS (90 D.U.)


0

10

20

30

40

50

60

0 50 100 150 200 250 300 350

Single Rate on a PMT (kHz)

N.

of

TC

PU

s

!KNCNPMTNTCPU = RCPU

RCPU SHitNACC = Drate HM#TCPU

Drate HM#TCPU = 10 GbpsSHit= 224 b

RCPU = 3 GHzNACC= 70

NPMT=8000NC=NACC=70

Number of Available Clock Cycles per CPU per TS:

Estimed number of necessary TCPU:

If the req. NC>NACC(i.e. trigger algo. is slow)

ADD TCPU!Tommaso Chiarusi RICAP 2009

Monitoring

Dispatcher for

TriDAS status

Dispatcher for

Physics Data

Trigger Time Window

transferred to EM

events written and

stored

Time Slices

tranfser to TCPU

TCPU

Via ControlHost: a Tag Controlled Data Dispatching [V. Maslenikov et al. CASPUR, http://afs.caspur.it/temp/ControlHost.pdf]


Multi purpose on-line Visualizer ( with D. Bonfigli - UniBo)

On(Off)-line Event Display ( with A. Riccardo - UniBo)


Conclusions“all data to shore” is a challenging BUT feasible strategy for km3 underwater !-telescope;scalable TriDAS architecture suppliyng high data-stream (up to 500 Gbps) is possible and affordable with the present technology;The NEMO Collaboration is completing a scaled TriDAS for the prototype Detection Unit.


G. Priede et al., Deep-Sea Research I 55 (2008) 1474–1483

scalable tridas for the nemo projectricap09.roma2.infn.it/slides/chiarusi_ricap2009r.pdf · n. of...

Documents