Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://www.eso.org/projects/dfs/papers/mess84/
Дата изменения: Wed Jun 10 17:55:22 1998
Дата индексирования: Sun Dec 23 03:07:04 2007
Кодировка:
Поисковые слова: m 8
|
|
|
Data Flow System
Overview |
|
Introduction
Astronomers are currently at an important cross-road in their
ability to gather information about the Universe. Firstly, computers
and astronomical detectors have simultaneously increased in
performance. Modern observatories using high-throughput, large-format
detectors for imaging and spectroscopy are capable of producing tens
of gigabytes of raw data every clear observing night. Although this
allows rapid progress on observing programmes, the data rate and
volume can easily saturate current network technologies and consume
the disk space typically found on most workstations. Secondly, large
telescopes in the 8 10-m class are going to be the workhorses of
optical and IR astronomy well into the next century. These telescopes
will be available to astronomers around the globe and competition for
these expensive resources will be fierce. Observing programmes
then have to be undertaken in the most efficient manner possible,
minimising the effects of the weather in order to complete programmes
in the shortest possible time.
Because of this "data deluge" and the need for maximum efficiency,
astronomers are faced with the realisation that perhaps the classical
observing cycle of trekking to the telescope and returning home with a
tape of unprocessed data is no longer the most efficient or practical
way to do astronomy. ESO astronomers stood at this crossroad and
considered the options carefully when designing the Science Operation
Plan for the VLT.
Operational Aspects
In the Science Operation Plan, ESO decided to enable the VLT to execute
observing programmes in a service mode. Astronomers can specify
programmes in a Phase I Phase II proposal process and the resulting
observations would be completed by ESO astronomers and technicians at
the VLT when the conditions are most favourable for that pro-
gramme. In this way, ESO astronomers would be as efficient and
competitive as possible in completing the types of frontier programmes
that will be common place on the VLT. Furthermore, ESO decided to
meet the challenge of the data volume potential of the VLT by
insisting that the VLT should be able to deliver calibrated data to
the astronomer with a well-defined accuracy. This has two immediate
benefits for the astronomer. Firstly the data processing and
storage necessary to calibrate the data is provided by ESO which helps
relieve the deluge significantly since most data reduction systems
expand raw data by factors of 2 to 5 before calibration is
complete. Secondly, by guaranteeing the calibration process to a
given accuracy, ESO has to continuously monitor the applicability of
calibration data and the performance of telescopes and instruments
over long periods of time. The quality control of the calibration
process will ensure the longterm usefulness of the data.
The VLT Data Flow System
Once ESO adopted the principles of service observations and delivery
of calibrated data for the VLT, it was obvious that the flow of data
must be managed from start to end of the observing process. The
design, development, operation, and maintenance of the VLT Data Flow
System (DFS) are done by the Data Management Division (DMD). This
software and hardware is responsible for the flow of scientific data
through the VLT from Proposal Entry to Archival Research.
DFS Concepts and Design
In early 1995, the Data Flow Working
Group was formed to formulate the requirements on the DFS following
the guidelines of the Science Operation Plan. The group consisted of
astronomers, instrumentation and software experts from several ESO
divisions. Discussions within the group focused on the
characterisation of the observing cycle from proposal entry to final
archive of the data. This cycle was broken down into component
processes and the outline of the DFS took shape.
A fundamental concept that holds the component processes together
is that of the Observation Block. This quantum of data in the
data flow system is an object which contains the essential target and
instrument information to define an observation of a single
object. An Observation Block is defined by either a classical observer
at the telescope or a service mode observer. Whether the observer
uses service-mode capabilities or not, the Observation Block is
intrinsic to the flow of data from place to place in the VLT
operation.
DFS Modules
Preliminary design documents describing the services offered by each
of the modules in the DFS and their interrelationships
were drawn up and distributed by the middle of 1995. The basic
modules in the DFS are:
- Programme Handling:
The
electronic submission of Phase 1 proposals. Preparation of a
6-month schedule for service and classical mode observing sessions.
- Observation Handling:
Phase II
proposals are prepared using software tools. Observation Blocks
are created for each target. Medium-Term Scheduler system prepares
a list of possible Observing Blocks to be executed for the next
few nights. Short-Term Scheduler systems select the next
Observation Block to be executedby the VLT Control System (VCS)
based on environment and configuration
data.
- Science Archive:
Provides
on-line catalogue services to observers. Stores raw data passed by
the VCS. Data distribution and archival research on VLT data.
- Pipeline:
Organise raw data
and calibration information for analysis. Schedule calibration
processing of data. Apply approved calibration recipes.
- Quality Control:
Maintain
calibration plan for a given instrument and mode. Compare
calibration data with simulators. Study long-term trends in
calibration data. Assess quality of calibrated data with respect
to standards.
The DFS is the scientific operating system of the VLT. The VLT can be
viewed as one machine with multiple components that interact. Unless
the components of the VLT machine work together in an efficient
manner, the scientific potential of the VLT will not be fully
realised. With the DFS as a science operating system, the VLT machine
can be scheduled and can make use of redundancy, pipeline and
aggregation of resources to maximise throughput. In the design of the
DFS, many opportunities exist to exploit the single machine vision
of the VLT.
DFS and VCS
The concept of the DFS arrived late in the process of designing and
implementing of the VCS. The VCS system is responsible for the control
of the VLT Unit Telescopes and their associated instruments. This
responsibility extends from low-level motor control to high-level
assessments of the technical quality of data and its movement from
instruments to storage. VCS was designed and partially
implemented before concepts like Observation Blocks were
introduced. It is therefore clear that the success of the DFS will
depend critically on the close collaboration of software and system
designers from DMD and the other ESO divisions. In September 1995,
a Data Flow Project Team was formed by the Data Management Division to
bring together astronomers and engineers from the VLT, Science and
Instrumentation Division to address the many interfaces issues opened
between the DFS and VCS. This Team has worked hard to crystallise the
critical system interface issues and a number of collaborative
software programmes are now under way between DMD and the VLT software
group. A review of the status of VCS and DFS design in
April 1996 pointed out the importance of the ongoing work of the Data
Flow Project Team.
|
The development of a close collaboration and coupling between VCS and
DFS has been aided and accelerated by the prototyping of DFS modules
and VCS on the NTT. In July 1996, the NTT took off the air for six
months in order to install VLT-specific hardware and software for
testing. From the beginning of 1997, the NTT acted like a single unit
telescope. Several modules of the DFS, including the Observation Block
System, have been tested extensively using SUSI as a trial instrument.
All modules of the DFS have been tested and and will modified within the
NTT prototyping environment before going on a VLT unit telescope.
Status of the DFS
There are two critical interfaces between DFS and VCS which allow the
basic operation of the DFS to proceed. The first allows Observation
Blocks to flow from the Short Term Scheduler into VCS. Observation
Block substructures call templated are interpreted by the Sequencer
within the VCS, allowing instrument operation and target acquisition
to proceed. A second interface allows completed science frames to be
stored on the OnLine Archive System (OLAS). Within OLAS, science
frames are recombined with associated technical data to supplement
the associated Observation Block for the observation.
In order to meet the timeline of the VLT, it is essential to bring DFS
components design to operational prototype as quickly as possible. To
achive this, a rapid prototyping process was identified using the
NTT. DFS prototypes have been commissioned and operated on the NTT in
a fully VLT compliant environment. DMD identified SUSI as a tarher
instrument on which to test the DFS. Prototypes for Phase II Proposal
Preparation, Calibration Pipeline, On-Line Archive and Scheduling
System at the NTT started in January 1997, after extensive testing
in December 1996 at ESO Headquarters and the NTT. For 1997, prototypes
for SOFI and EMMI are planned.