Transcript CTA as

Operations
Report of the_-----------_____
Robotics
‘Computing and Readout
Readout’
Working Group
presented by
Adrian Biland / ETH Zurich
CTA Meeting
Paris, 2. March 2007
Active WG Members
- Angelo Antonelli (REM)
- Adrian Biland (MAGIC)
- Thomas Bretz (MAGIC)
- Stefano Covino (REM)
- Andrii Neronov (ISDC)
- Nicolas Produit (ISDC)
- Ullrich Schwanke (HESS)
- Christian Stegmann(HESS)
- Roland Walter (ISDC)
(several
on mailing
list)of
existsmore
internal
draft
Paris, 2.3.07
A.Biland, ETHZ
a TDR-like paper
IT working group report
2
two fundamentally different
approaches how to operate
- collaboration mode (as in particle physics)
- observatory mode (usual in astronomy)
main differences (simplified, exist mixed modes):
funding:
manpower:
(salaries)
Paris, 2.3.07
collaboration
observatory
Inst. ask fund. Agency;
build part of experim.
PhD students (ab)used
as workhorses for
construction and
operation
Agencies pay to
‘Consortium’
construction:
salaries incl. in cost
operation:
hire staff personal
A.Biland, ETHZ
IT working group report
3
proposals
from collaborators
from anybody -> PI
data owner
authorlist
full collaboration
(huge authorlists)
PI ( and CoIs)
(short authorlists)
data
availability
kept internally
(often lost at the end)
public domain (delayed)
==> ‘data mining’
‘personal
dedication’
start as PhD and stay
until retirement (LHC)
- have an idea
- check data archive
- write paper
- get Nobel Prize
(without any contribution
done to the experiment)
extreme cases
‘My Experiment’
But usually experiment
renowned not the PI: e.g.
‘HUBBLE found...’
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
4
doing physics
Particle Physics: (usually) clean, well defined environment
clean, self-consistent theories to check==>predictions
dedicated, 'simple' experiments to test predictions
do EXPERIMENTS ==> collaboration mode ok
Astronomy: no control over environment and 'setup of experiment'
no clean theory; fundamental stuff obscured by complicated
standard-physics processes
usually need many sources and/or many observatories (MWL)
to get conclusive answers
OBSERVATIONS ==> observatory mode, data mining (for MWL)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
5
doing physics
What about CTA ? We are in a paradigm-shift:
WHIPPLE, HEGRA, CAT, .... : 'source hunting'
- proof of concept
- invent new fundamental techniques
- new source ==> new conference
Collaboration mode is ok
H.E.S.S., MAGIC, ...: 'proof VHE important part of Astronomy'
- mature technology
- surprising richness of VHE sky
- impressive results; few conclusive answers
- MWL becoming more and more important
===> must incorporate external experts
Collaboration mode getting difficult (authorlists !!!)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
6
doing physics
CTA: 'understand physics' (hopefully)
- expect ~1000 sources (cannot use a PhD per source)
- need automatic data (pre)processing
- can do statistical analysis
[ compare astrophysics:
learned a lot from statistics of Hertzsprung-Russel Diagram ... ]
- MWL becoming extremely important
- for steady/periodic sources: data mining
- for transients: additionally dedicated MWL campaigns
Final goal: UNDERSTANDING PHYSICS
==> need to incorporate brightest minds
(not all of them will be willing to dedicate themself several years
to an experiment ...)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
7
doing physics
==> CTA better to be operated (part time)
as Observatory (allow guest observers)
and allow data mining (public data access)
details to be discussed/agreed:
- how much time for guest observers
- how long delay until data become public domain
can make the difference if CTA will be seen as
obscure corner or major pillar of astronomy ....
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
8
CTA as (open) Observatory
- Need well defined procedure how to submit
and process proposals
- Existence and well defined access to data and
analysis programs
...
What is more efficient:
centralized or de-centralized structures ???
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
9
‘Array Control Center’ (onsite)
tasks:
- monitor operation of CTA (goal: automatic/robotic operation;
but having >>10 (100?) telescopes, there will be hardware problems
- ensure safety (nobody within array during operation; what about
power failures at sunrise; ....)
- buffer raw-date until shipped to data center (even in
case of fast internet connection, we must foresee a buffer in case of
problems)
- monitor Quick-look analysis (onsite analysis)
- ....
Most can be done by local technicians
(but if we want to send out alarms, need verification by experts)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
10
Operations (center?)
- submission and handling of proposals
- plan operation of CTA; scheduling
- handle incoming ToO Alarms [GRB: directly to ArrayCtrl]
- control operation of CTA (automatic/robotic operation ==>
can also work if there is some downtime in communication)
- control hardware status of CTA (slow control level)
- ...
Needs CTA hardware/physics experts (available/on call)
(could be 12 timezones away ==> no night-shifts ?!)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
11
Data (center?)
exist different possible implementations;
extreme cases:
- data-repository
- ‘science support center’
For (open) Observatory:
It is less important to have a place where
disks are located, but to have a dedicated
contact point for inexperienced users
(to be defined on what ‘luxury level’ this has to be:
users get raw data vs. users get ‘final plots’)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
12
Data (center?)
- receive and store data from CTA
- calibration (==> check hardware status)
- automatic (pre)processing of data
- archive data on different stages:
raw, compressed, preprocessed, photon-list, ...(?)
- ensure availability of data to predefined groups
PI, CTA consortium, public domain
- make available (and improve) standard analysis tools
- offer data analysis training sessions/schools
- ...
Needs CTA Analysis Experts
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
13
General Remark:
CTA staff physicists shall be allowed/
encouraged to spend reasonable part
of time to do their own physics within
CTA (as PI or CoI ...)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
14
Design Studies
On next slides,
topics for design studies are marked in blue
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
15
Towards CTA Standard Analysis
0) adapt H.E.S.S./MAGIC software to analyze CTA MC
for the Hardware Design Studies (partially done)
1) define needs
-underlying data format (FITS, root, ...)
-interfaces to external data (for MWL analysis)
-what amount of RAW data must be archived
for re-analysis (FADC slices? only Pixels above threshold?)
[might have to archive several PBytes/year]
- ....
2) tools survey (what exists and can be re-used)
3) start programing (rather early to be ready for first telescope!)
package must be useable by non-experts ==> ‘KISS’
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
16
Towards MC Mass Production
In hardware design studies (and during operation),
need huge amount of MC ==> urgent
-tools survey
-optimize/speedup programs
GRID:
-exists (most probably mature soon)
-EU spent €€€€€€€€ ==> success by definition
-at the moment, huge amount of unused CPU available
MC easiest to use GRID, can most profit from GRID
==> concentrate GRID activities in MC package
Analysis software shall be GRID aware, but not rely on it
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
17
Towards Array Ctrl: ‘robotic’
too complicated system to make fully robotic
hardware must be very(!) reliable
software must be rather simple
(no bells and whistles...)
limited experience, need test environment
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
18
Towards Array Ctrl: Slow Ctrl
-centralized approach:
powerful central machine controls individual (dumb)
telescopes; knows always everything about
anything....
-distributed approach:
each ‘intelligent’ telescope runs independent;
Central Control just distributes tasks (e.g. schedule
changes) and receives status info (but can always
request any infos)
-mixed mode
Design study: find optimal solution ...
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
19
Towards Array Ctrl: Trigger
-most simple approach:
-each telescope has its own local-trigger
-multi-telescope trigger just combines next neighbors
local-trigger information
-very ambitious approach:
combine adjacent telescopes on the pixel level
(technically feasible ? any advantage ???)
Design study: find optimal solution ...
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
20
Towards Array Ctrl: DAQ
-centralized approach:
CC receives all raw-data and writes combined ‘CTA
events’ containing several telescopes
-distributed approach:
each telescope writes its own data stream (including
trigger-info) to local or central file server; combining
data of adjacent telescopes done at analysis stage
Design study: find optimal solution ...
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
21
Towards Operation (center)
probably only few CTA specific requirements
==> need only tools survey (and take decision)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
22
Towards Data (center)
- define needs <== experts from Cherenkov and Astro
- tools survey
- use HEGRA data as playground to proof
feasibility of approach
possible extension:
[midterm]: archive H.E.S.S./ MAGIC data in
‘universal’ format for extended tests
[longterm]: allow data mining on old H.E.S.S./ MAGIC
very political decisions...; important signal to astro-community...
also analysis package should be tested on old
data
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
23
Towards Data (center)
is the interface to the astro-community
(data mining !)
basic design might not be crucial for us,
but it should ‘please’ the other users ...
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
24
Summary
-Exists lot of experience in all needed fields
-Important to combine and synchronize knowledge
-Several decisions to come out of design studies;
find ‘optimal’ solutions (probably now show-stopper:
‘wrong’ solution will also work, but less efficient and more
expensive)
-Some tasks will need rather long time between
design study and final product (e.g. full analysis
package); to be ready in time, these design
studies must be finished rather soon so that
implementation can start ... ( e-center call in 2008 ???)
Paris, 2.3.07
A.Biland, ETHZ
IT working group report
25