1. KLOE @ DAFNE Computing and Network Maria.Lorenza.Ferrer@lnf.infn.it
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

2. INDEX DAQ and OFFLINE requirements Computing and Network
Physics at KLOE
The KLOE detector
DAQ and OFFLINE requirements
Computing and Network
Design Strategy
Achievements
Present Status
Archiving
Hardware and Software
Database
Software
Present status
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

3.    Physics at a f - Factory At full Luminosity 51032 cm -2 s -1
8.5x KLKS
1.2x K+K-
~ f0 g,a0 g
2.5x108 , 2.5x hg, h’ g
At full Luminosity 51032 cm -2 s -1
in 1 Year 
Measure all the relevant CP CPT violation parameters from INTERFEROMETRY and DOUBLE ratio 
Kaon form factors, Ks rare decay and Ks semileptonic
asimmetry (never measured) 
Radiative f decay  investigation of the f0 , a0 nature & precise determination
of BR(f h’g)/ BR(f hg)
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

4. The double ratio method @ KLOE
The traditional way to measure direct CP violation is through the double ratio:
R = = 1+6(e/e)
where:
N(KLp+p-)/N(KSp+p-) N(KLp0p0)/N(KSp0p0)
observed
measured
I(l) = decay intensity
g(l-l) = experimental resolution
NKK= Lint  sfBR(fKSKL)
NKK & etag drop out identically in the double ratio
@KLOE we are aiming for:
d  (e/e) ~ dR ~ few 10-4
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

5. Statistics
The double ratio
The goal for KLOE is to measure e/e to ~10-4
Since: R 1+6 e/e 
4106 KL  p0p0 events are required, corresponding, with KLOE FVs and tags efficiencies to ~ 2 years of data taking @ L=51032cm-2s  41010f
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

6. With a Luminosity L = 5x1031 cm-2s-1
FIRST PHYSICS
With a Luminosity L = 5x1031 cm-2s-1
Our first goal is to collect 100 pb-1 of data ( 3x108 F), which should allow the following physics results:
Measurement of  (e/e) to 10-3 statistical accuracy;
Measurement of K3 form factors;
Meas. of BR(f  f0g  p0p0g) to 5 better accuracy;
Confirmation of f  a0g  hp0g and measurement of its BR to  5 better accuracy.
Now 4x1030
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

7. Typical e+e - general purpose detector
The KLOE Detector
KLOE
Typical e+e - general purpose detector
~5 m diameter & ~4 m length
Beryllium Beam pipe (radius >16 lS)
Two Quadrupole triplets Calorimeters (32 PMs)
Helium Drift Chamber (12,582 Sense Wires)
Lead-Scintillating Fiber Calorimeter (4,880 PMs)
Superconducting Coil of 0.6 T
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

8. The Electromagnetic Calorimeter
Requirements
 Determine the vertex of KL,S neutral decays with an accuracy of few mm
 Have an high discriminant power on neutral modes K02p0 and K03p0
 Provide a fast and unbiased First Level Trigger
 Provide useful information for particle identification
1.2 mm
Lead
1.35 mm
1.0 mm
52.5 cm
15X0
450 cm
Solution
Fine sampling lead/scintillating fibers calorimeter
Volume Ratio Fiber:Lead 50:50
Energy sampling fraction: 13 %
Good energy resolution (~5% / Ö E (GeV) )
Fully efficient in the range MeV
Excellent time resolution (~70ps / Ö E (GeV) )
Determination of g conversion point with ~1cm accuracy
Hermetic (rejection of ~10-4 on KL ® 6g )
Fast triggering response to suppress the 20 KHz Bhabha rate
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

9. The Drift Chamber Requirements Solution
 High and uniform track reconstruction efficiency
 Determine the KL,S vertex with an accuracy of 200 mm x 1mm
 Good momentum resolution (dp/p 0.5% ) for low momentum tracks
 Transparent to low energy g (down to 20 MeV) and KL,S regeneration
Solution
High homogeneity, isotropy, large volume (f ~4 m, L~3.3 m, wires).
All Stereo layers with constant stereo drop =1.5 cm,  =  (60150) mrad
12 layers of inner 2x2 cm2 cells  46 layers of outer 3x3 cm2 cells
Helium (90%He-10%iC4H10 ) gas mixture
Al(Ag) 80 mm field wires, W(Au) 25 mm sense wires, X0(gas+wires)=900 m
Very thin walls: mechanical structure entirely in C-fiber/epoxy (  0.1 X0 )
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

10. DAQ handles ~ 23000 FEE channels
on 2.5 kHz f + 5 kHz bckg
Signal conversion/digitization in 2 ms
Dead time 2.2 ms
Bandwidth: ~ 50 Mbytes/s (5 Kbyte/ev.)
Storage: 200 Tbyte/y
Fully tested with peak rates of 10 kHz in
multibunches mode.
Tested at maximum required throughput
using no zero suppressed calorimeter data
Connection to offline world through
a switched Gbit/FastEther network
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

11. KLOE Data Taking: Short history
Apr 14th: First collisions
Apr 14-May10
Stable data taking in single
bunch mode:
Φ line-shape scan (30 nb-1)
From May 10th collisions in
multibunch mode started. Few weeks
of parasitic running (30 nb-1)
July 30th -Aug 8th our first period of “continuous” data taking.
data taking duty cycle > 95 %!!
200 nb-1 collected
Parasitic running now
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

12. ONLINE & HISTORY FILES ACCESS
ROOT
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

13. COMPUTING & NETWORK requirements
Acquire up to 50 MB/s (10 kHz) in about 1 hour runs (200 GB & files)
Online monitoring for
DANE conditions (beam shape, luminosity, background)
Detector conditions (hot/dead channels, efficiencies)
Need fast event selection and reconstruction of event samples
600 SpecInt95
Archive raw data on tape but keep data on disk to allow for first reconstruction
Disk space > 10 TB
Offline reconstruction at 10 kHz and concurrent re-processing /analysis
Archive/recall bandwidth up to 200 MB/s, 6000 SpecInt95
Network
Efficient TCP/IP transport
Efficient access to remote disk (NFS)
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

14. SOFTWARE DESIGN Multi-platform environment Online
in-house development starting from scratch (C) (no OO)
experts people for initial design
Offline
in-house development (Fortran)
CERN & FERMILAB tool
Archiving/Backup/HSM
First tests on Legato/Fatmen/Shift
ADSM adopted
Database
HepDB for calibrations constants
IBM DB2 RDBMS for detector maps, run conditions,
acquired /analyzed file status
Move to commercial software when it exists at convenient price
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

15. COMPUTING & NETWORK acquisitions plans
< 1997
Online development using available CPUs for evaluation
FDDI chosen for Detector and Online Farm interconnections (switch available)
Online & Offline development on multi-platform environment
1997
CPUs in VME crates for detector readout
1/10 Online Farm
1998
Full Online Computing power
1/10 Offline Computing power
Tape Library & 100 TB tapes
Network switches
Complete Offline Computing power, more tape drives, up to 1 PB (?) on tape
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

16. COMPUTER SELECTION CRITERIA
=> General requirements about OS & Compilers (a given # of Spec)
=> Network capabilities
=> Reduced # of nodes (no PCs) due to
- short manpower
- high reliability required to simplify management
=> Final choice according with
- technical constraints
- available expertise & price
DEC ALPHA based for VME crates
IBM servers for Online
SUN servers for Disk serving and Offline
Xylan switches for Networking
IBM library for Archiving
‘97 (7 servers)
‘98 (12 servers)
installed
in ‘99
All running now
Tapes ( 8 TB + 5 TB, ESP in Nov)
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

17. KLOE Computing Network
FDDI/Ether.
Hubs
Gbit and Fast
Ether. switches
Xterminals
@ Control
room
Eth.
FDDI
Detector
Readout &
Controls
Fast-Ethern.
Offline
Farm
Gbit-Ethern.
Tape & Disk
servers
FDDI
Switch
FDDI
Eth.
Gbit-Ethern.
Fast-Ethern.
Run Control
Slow Control
servers
Online Farm
@ Computing
room
LNF
Network
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

18. XYLAN Switches CONFIGURATION
DETECTED
SET
Slot/
Intf
Auto-
negotiate
Line
Speed
Duplex
Mode
Line
Speed
Duplex
Mode
Xylan #1
3/ enabled FULL-D auto auto SUN
3/ disabled HALF-D half-d NCD Xterm
3/ enabled FULL-D auto auto TEK Xterm
4/ enabled FULL-D full-d SUN
5/ disabled FULL-D full-d IBM
5/ disabled FULL-D full-d
6/ disabled FULL-D full-d IBM
Xylan #2
2/ disabled FULL-D full-d
2/ disabled FULL-D full-d IBM
3/ disabled FULL-D full-d DEC
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

19. ARCHIVING & ADSM PERFORMANCES
26 can be installed
KLOE IBM 3494 Tape Library with 5,500 slots
redundancy, high reliability, 6 (=>12) drives 3590E 20 (=>40) GB
Up to
30 MB/s
disk <=> tape
each server
Two servers
1/2 library each
H50 4-way
1GB 72GBdisk
H50 4-way
1GB 72GBdisk
Switched
Gbit-Ethernet -
Recalled areas
NFS v3 exported
Up to
9 MB/s
disk <=> tape
SUN3500
2-way 1GB
SUN3500
2-way 1GB
SUN
500GB
Working areas
NFS v3 exported
Offline Farm
Online Farm
350 GB
Switched Fast Ethernet
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

20. ADSM: BACKUP & HSM KLOE IBM 3494 Tape Library with 5,500 slots
redundancy, high reliability, 6 (=>12) drives 3590E 20 (=>40) GB
H50 4-way
1GB 72GBdisk
H50 4-way
1GB 72GBdisk
HSM
run &
DANE
conditions
Scheduled
ADSM
backup
Online
backup
using
ADSM policies
DB server
active
DB server
standby
RUN server
FARM
ADSM clients
Mirrored OS
& Database
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

21. ADSM: MULTISERVER Using server-to-server connection
KLOE IBM 3494 Tape Library with 5,500 slots
redundancy, high reliability, 6 (=>12) drives 3590E 20 (=>40) GB
Import
from node
(2)
11.5 MB/s
H50 4-way
1GB 72GBdisk
H50 4-way
1GB 72GBdisk
Two phases
Using server-to-server
connection
Switched
Fast-Ethernet
Migration of 11,000 files
(1600 GB) in 60.5 hours
<==> 7.3 MB/s
F50 4-ways
Export
to node
(1)
LNF IBM 3494 Tape Library with 220 slots
2 drives GB
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

22. DATABASE SYSTEM DAQ configuration run conditions
acquisition electronics
configuration
run logger
general description of runs
mapping between
detector & DAQ
channels
file logger
description & attributes of raw
& reconstructed data files
=> granularity at run level
slow control logger
detector and DANE
parameters
2.5 KB/
run
=> several rows per run
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

23. DATABASE ACCESS IBM DB2 RDBMS
very well supported in multiplatform environment
well interfaced with ADSM
dedicated tools
simple client-server protocol
multithreaded (up to 256 concurrent connections)
based on SQL commands
preferred over DB2 client-server solution
less resources & administration needed
more efficient using caches on server
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

24. Tape Library/recalled areas
DATABASE ACCESS: a flowchart
Offline farm
Tape Library/recalled areas
VME-ROCK
Rec file
Datarec DB
Collector
Raw file
VME
Recorder DB
Sender
L3 filter
Receiver
Raw file
Calib
Builder
Recorder
Online farm
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

25. DATABASE ACCESS TCL/TK
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

26. DATABASE ACCESS
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

27. ARCHIVING & RECALLING FILES - the KLOE HSM
Raw & reconstructed data files are archived by dedicated daemons
when closed
ONLY if they are “declared” in the Database
Files in use are “marked” in the Database
Files are deleted from disk by dedicated daemons
disk space is required
not in use but already analyzed, ...
Files “analyzed” with a given program version can not been
re-archived (program version should change)
recall raw s11 1 “(run_nr between 9601 and 9605) and (stream_code = ‘L3BHA’) and (…) ….”
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

28. DATABASE ACCESS List_runs -disk tail 20 List_raws 9859
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

29. DISK SPACE ORGANIZATION
Using ADSM -API
specialized daemons
on ADSM and DB2
servers.
Offline
servers
Reconst.
Automatically archived
recalled
Retrieved at request
User area
Backup vs HSM
In progress
Online
servers
Resident daemon for
disk space management
Reconst.
Automatically archived
Raw files
From DAQ & filters
Run &
slow &
DANE
controls
History
files.
HSM
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

30. We have successfully tested
Conclusions
We have successfully tested
Our detector performances
Our trigger and DAQ reliability
Our online monitoring
Our archiving/backup/HSM & Database systems
Our reconstruction and filtering procedures
Using powerful CPU servers that offer high reliability at relatively low cost
short manpower but > 1700 SpecFp95 running
Data reconstructed at 3-4 kHz
Stable high bandwidth network
Commercial software for Tape library management and Database
NIM software for IBM software installation, in-house utilities for system and production management
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

31. Candidate
    p+p- 
M =498 MeV
M’=908 MeV E
= 57 MeV

32. The Electromagnetic Calorimeter: Time performances
s(mip)  300 ps
OutsideInside
& Inside  Outside
muon nicely recognized
(used to rejet comsic rays)
Outgoing m T5 5 4 3 2 1
Ingoing m T1 T1
T1-T5 (ns) T1 b
b = L/(T1-T2) T1
Fit b vs p distribution
Check the m mass L T2
p (MeV/c) (DC)
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

33. Bhabha momentum resolution
The Drift Chamber: Performances on bhabha and F
Bhabha momentum resolution
Ks+-
s ~ 1 MeV
s/ P ~ 0.4%
Mks(MeV)
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

34. LDAFNE = 51032 cm -2 s -1 DAFNE: Design strategy
L VEPP-2M = 51030 cm-2 s-1
“conservative” single bunch approach:
L0 (single bunch) ~ L VEPP-2M
but large number of bunches
L DAFNE = nbunches  L0
nbunches 120 Þ Two Separate Rings
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)

35. Permanent Magnet QUADRUPOLES
QCAL
45 cm
0.5 mm Be 9°
Permanent Magnet QUADRUPOLES
Instrumenting the Quads
improves the rejection of
KL p0p0p0 by a factor of 5
Timing has been set. It helps to reject machine background
CERN HEP-CCC Nov 12’ Maria Lorenza Ferrer (LNF-INFN)