1. Emanuele Leonardi PADME General Meeting - LNF 17-18 January 2017
On-line Computing
Emanuele Leonardi
PADME General Meeting - LNF January 2017

2. PADME on-line computing model
Target
32ch
FADC
Spectrometer Veto
192 ch
FADC
High Energy Positron
Veto 32ch
FADC
ECAL
616 ch
FADC
SAC
49 ch
FADC
TPix
65536 ch
TBD
L1 DAQ
1 process per DAQ board
L0 DAQ
1 process per DAQ board
ADC zero suppression (?)
Trigger Distribution System
BTF Beam
Cosmics
SW/Random
L1 DAQ
Event build
Temporary disk buffer
Neutral Filter (Inv) 1 or more ECAL clusters
Charged Filter (Vis) 2 or more tracks
RAW data
Trigger signal
Data flow
Central Data Recording Facility
CNAF (+LNF?)
PADME experiment site
PADME on-line computing model
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing 2

3. DAQ data rates
Detector
Channels
ADC boards
DAQ in
MB/s
DAQ out
Raw event
KB/event
Target 32 1
2.6
1.26
24.6
Calorimeter
616 20
52.8
2.70
SAC 49 2
5.3
1.08
21.1
E veto 96 3
7.9
0.60
11.6
P veto
1.31
25.6
HEP veto
1.15
22.5
TimePix
65536 -
18.8
1.50
29.3
Total
919 30
97.9
9.60
187.5
Output estimates assume that zero-suppression is active,
include trigger ADC samples (25%) and a 1% autopass stream (10%),
and do not include ROOT data-compression factor (testbeam: ≈50%)
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

4. E. Leonardi - PADME G.M. - On-line Computing
L0 DAQ system
Data acquisition + Zero suppression (?)
2 DAQ servers directly connected to ADC boards
Each server will mount 2 CAEN A3818 optical boards
CAEN will provide 4 A3818 with the ADC boards
Current server: HP Proliant DL560 gen 8 (2015)
Now available from CONSIP: HP Proliant DL560 gen 9
Possible alternative: DELL PowerEdge R630
Cheaper (-30%) but less performant/expandable
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

5. E. Leonardi - PADME G.M. - On-line Computing
L1 DAQ system
DAQ data buffer + Event build + Event filters
I/O bound
in: L0 data,
local: DAQ→RAW
out: RAW to CNAF/LNF
Storage requirements depend on zero suppression
Zero suppression on-line → 1-2 TB/day
Zero suppression off-line → ≈ 10 TB/day
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

6. Server Dell PowerEdge R730xd 2 CPU Xeon E5-2630L v3 8 core 64 GB RDIMM
2 x 300 GB RAID1 (system) + 22 x 1 TB RAID5 (storage)
2 x 10 Gbps Ethernet
2 servers → 2-10 days of data buffer
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

7. Local Network Infrastructure
Ethernet switch 48 × 1Gbps + 4 × 10Gbps (e.g. HPE 5130)
8 × 1Gbps: L0 servers (4x1Gbps trunking if needed)
2 × 10Gbps: L1 servers
1 × 10Gbps: uplink to LNF-IT
n × 1Gbps: other services (DCS, Trigger, HV, …)
Will buy 2 switches for redundancy/backup
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

8. E. Leonardi - PADME G.M. - On-line Computing
Other infrastructure
The whole online system (servers+switches) fits in 1 rack
Airflow/air conditioning must be guaranteed (should be easy)
UPS?
Network cabling depends on the positioning of L0/L1 systems
In/near experimental area
Need to setup 10Gbps fiber link to LNF-IT
Connectivity to Trigger system
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

9. 2017 buying list 2 L0 servers (HP Proliant or Dell PowerEdge) 10-15K€
2 L1 servers (Dell PowerEdge) K€
2 Network Switches (HPE 5130) K€
Rack & cabling K€
All HW to be ordered before summer
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing

10. E. Leonardi - PADME G.M. - On-line Computing
DAQ software update
Improved zero-suppression algorithm
Channel is accepted if RMS is large
Total number of events per run is written to DB
New web interface to Padme DB
17/01/2017
E. Leonardi - PADME G.M. - On-line Computing