Online%20Monitoring%20System%20at%20KLOE - PowerPoint PPT Presentation

Title: Online%20Monitoring%20System%20at%20KLOE


1
Online Monitoring System at KLOE
CHEP 2000 Padova, 7-11 February 2000

• Alessandra Doria INFN - Napoli
• for the KLOE collaboration

NAPOLI
2
The monitoring system
To be used by experts and by shift takers has to
be flexible
An heterogeneous set of tasks perform on online
data different kinds of monitoring (detector
response, trigger, data taking, beam parameters,
data quality ...).

• Monitoring is managed by Run Control, as DAQ, but
  the two system are independent.
• The monitoring tasks get data from the online
  data stream
• no interaction with DAQ processes
• no reducing DAQ performances
• minimize data transfer.

3
The monitoring system
To be used by experts and by shift takers has to
be flexible

• The development of the monitoring tasks has
  involved experts from all KLOE subgroups.
• The tasks are implemented either in C, C,
  Fortran
• different graphic tools used for display of
  results

• Spying A common mechanism to access online data
  has been designed and implemented.
• get event data as soon as they are available in
  the online data stream.
• do not disturb main data flow.

4
Online farm data flow
Packets of sub-events from DAQ chains
Monitoring nodes
Receiver
Online Farm node
Builder
Recorder
5
Spy daemon
To be used by experts and by shift takers has to
be flexible
Monitoring tasks may run locally on a farm node,
spying on events directly from the local buffer,
or can be distributed on remote nodes. A spy
daemon runs in each farm node as event server.
Packets of events are read from the local buffer
and sent on TCP/IP to remote monitoring tasks.
Spy daemon is a standard KLOE process, managed
by Run Control in the same way as the DAQ
processes. It runs with lower priority then the
other farm processes.
6
Spy daemon
To be used by experts and by shift takers has to
be flexible

• Spy daemon creates a new thread for each client
  connection.
• Three data serving modes may be chosen by the
  clients when the connection is established
• pop - the client requests a new packet as soon
  as processing of the previous one is completed.
• pop in advance - the client issues a new request
  as soon as a packet is received.
• push - spy daemon serves asynchronously, until
  the TCP buffer is full.

7
Online Event Selection
To calculate some physics parameters, only
special types of events are needed fast event
classification allows to select cosmic rays, ee-
and g g events. Event filters fill output
buffers with selected events. Other monitoring
tasks can spy on filtered buffers.
Recorder
Builder
ee-
Event Filter
g g
Recorder
Recorder
8
Trigger Monitor
Filters themselves may monitor physics
quantities. Trigger monitor selects events
(Bhabha and Phi decays) on the bases of trigger
data. It calculates DAQ rates, trigger dead time,
rate of Phi candidates, average and integrated
luminosity. Monitored parameters are displayed in
simple a tcl/tk graphic interface.
9
Histogram Monitoring

• Thousands of histograms filled during data taking
  for monitoring of detector readout, FEE and
  trigger
• client-server system, where the server is the
  central histogram producer and the clients allow
  histogram presentation
• Both server and client based on ROOT framework.

Spy daemon
Kbrowser
Kserver
Kbrowser
10
Histogram producer and browser

• KLOE server
• histograms organized in a tree of directories
  with 5 main branches Ecal, Drift Chamber, Qcal,
  Trigger, FEE.
• Can save current histograms into ROOT file.

• KLOE browser
• gets by Kserver the histogram directory tree
• allows the user to browse through the
  directories
• when a histogram is selected, requests updated
  data to Kserver and displays it
• when requested, gets updated contents of all the
  histograms in a directory and displays them.

11
Event Display and Physics Monitor

• Full event reconstruction with Analysis_Control
• (Fortran package used for offline).
• Specialized I/O module for spying .
• Physmon
• Spying on filtered streams, calculate physics
  quantities.
• Stores mean values and produces HBOOK histogram
  file when enough statistics is available. A ROOT
  based presenter converts HBOOK into ROOT format
  and displays histograms.
• Event Display
• OnX (from LAL, Orsay) objects for GUI and 3D
  graphics.

12
KID library
The KID (KLOE integrated data flow) library
contains all the C functions to get data from
different data sources. Local event buffers,
remote spying, raw data files on disk or archived
files may transparently be chosen as input for
all monitoring tasks. The same tasks are used for
online monitoring and for offline raw data
analysis.
13
Task distribution
Monitoring tasks have different requirements as
to event processing rate and CPU occupation.
14
Conclusions

• The spying method proved to be powerful and
  flexible monitoring tasks are connected to the
  online system preserving DAQ performances.
• Online filters to produce selected event stream
  turned out to be very useful.
• The possibility to merge input events from
  different buffers is foreseen.
• The configuration of the system, managed by Run
  Control, can be easily modified according to the
  evolution of requirements.