eSTAR: Distributed peertopeer scheduling

1
eSTAR Distributed peer-to-peer scheduling

• Allan, T. Naylor, E.S. Saunders
• School of Physics, University of Exeter, Stocker
  Road,Exeter, U.K.

2
The eSTAR concept

• Its an ad-hoc peer-to-peer network of
  heterogeneous resources utilising a collaborative
  agent paradigm for decision making
• But the aim is to disguise complexity, users hate
  complexity

3
eSTAR in a nutshell
4
What is an agent anyway?

• An agent is just software not magic
• Loosely, an agent is a computational entity which
• Acts on behalf of another entity in an autonomous
  fashion
• Performs its actions with some level of
  proactively and/or responsiveness
• Exhibits some level of the key attributes of
  learning, co-operation and mobility

5
Agents as architecture

• From a developers perspective there are five
  major trends which are evident from the history
  of computing.
• These are,
• ubiquity
• interconnection
• intelligence
• delegation
• human-orientation
• Agent architectures are the next paradigm shift
  following these trends

6
Multi-agent systems

• A multi-agent system is one that consists of a
  number of agents, which interact with one another
• limited intelligence
• emerging complexity
• cooperate, coordinate negotiate
• In the most general case, agents will be acting
  on behalf of users with different goals and
  motivations.

7
Peer-to-Peer

• Agents operate in a peer-to-peer manner and can
  make use of these interconnections between people
  and data.
• We tend to impose hierarchical structures on
  systems where it is not appropriate because it
  lets us grasp the system as a whole.
• Carrying out intelligent resource discovery could
  mean that your agent looks to your collaborators
  agent for data and expertise before it looks to
  central sources.

8
How does it work?
CREDIT Liverpool John Moores University
User Agent
Embedded Agent
UI
The Grid
9
Observing with eSTAR

• Importantly, the users doing science dont have
  to care which telescope takes their observation.
• observe at the best telescope
• optimally schedule the observations
• hand-off between distributed telescopes

10
Scheduling with eSTAR

• Complicated telescope schedules are built by
  humans using a multi-pass approach
• traveling salesman problem
• However by allowing access to your telescope you
  do not (have to) give up local control
• local scheduling at telescope
• distributed scheduling through the agent software

11
eSTAR as a meta-network

• eSTAR is not Robonet-1.0
• schedules observations onto the LT, FTN FTS
• eSTAR is not JACH
• schedules observations onto UKIRT JCMT
• generates event notifications
• eSTAR is not the HTN
• autonomous decision making
• distributed scheduling

12
The eSTAR network
Embedded Agents
Embedded Agents

• GRB rapid follow-up programme
• Search for exo-planets

13
eSTAR and UKIRT

• All aspects of an observation programme at the
  JAC are either software readable or software
  controllable.
• To the agent its irrelevant thats there is a
  human in the loop.
• Agent software has now been running successfully
  for a year on UKIRT doing GRB follow-up.

14
GRB 050716

• Successfully responded to the SWIFT XRT position
  alert and queued observations into the UKIRT
  queue by 48 seconds.
• Time to target should be around 5 minutes, or
  less.

The candidate afterglow is indicated by red bars,
and the XRT error circle in green.
15
eSTAR and Robonet-1.0

• Searching for extra-solar planets using
  gravitational microlensing events in the galactic
  bulge.
• About to move from queued to live scheduling of
  observations.
• Live anomaly detection and autonomous follow-up.

CREDIT Liverpool John Moores University
16
eSTAR and TALONS

• TALONS system very different architecturally than
  eSTAR
• A more traditional hierarchical architecture with
  a central node
• Gateway now operational
• event notification
• observing requests

CREDIT Los Alamos National Laboratories
17
Reacting to events

• Part of the emerging global event network
• GRB follow-up triggered by Swift alerts (via GCN)
• Events from the TALONS gateway will be feed back
  into the GRB programme on UKIRT
• Microlensing first look triggered by OGLE EWS
• Native event streams
• Robonet-1.0
• UKIRT (and eventually JCMT?)

18
VOEvent

• ltWhogt    Author Identification
• ltWhatgt    Event Characterization
• ltWhereWhengt    Space-Time Coordinates
• ltHowgt    Instrument Configuration
• ltWhygt    Initial Scientific Assessment 
• ltCitationsgt    Follow-up Observations
• ltDescriptiongt    Human Oriented Content
• ltReferencegt    External Content

19
The eSTAR network
20
VOEvent Network
CREDIT Roy Williams, Rob Seaman, Alasdair Allan,
Andrew Drake, Robert White, Matthew Graham,
Philip Warner
21
First light

• The first ever autonomous observation triggered
  by a OGLE EWS VOEvent message took place on the
  15th June 2006 around 1800 BST.
• event picked up by remote client
• passed to exo-planet user agent
• observations queued on Robonet-1.0
• We have therefore, for the first time ever,
  closed the loop for autonomous observations based
  on the VO's new VOEvent event messaging standard.

22
User buy-in

• 92 lines of Perl, including
• ACK messages
• IAMALIVE messages
• error handling
• re-connection
• message parsing

23
RSS feeds

• eSTAR native feed (includes OGLE III feed)
• http//www.estar.org.uk/voevent/eSTAR/eSTAR.rdf
• RAPTOR/TALONS feed (includes GCN translation)
• http//www.estar.org.uk/voevent/RAPTOR/RAPTOR.rdf
• Caltech feed (re-published GCN translation)
• http//www.estar.org.uk/voevent/Caltech/Caltech.rd
  f

See the eSTAR website http//www.estar.org.uk/
24
Where now for eSTAR?

• Battle hardened systems
• software designed for survivability and
  self-healing
• Ubiquitous computing, or astronomy everywhere
• prototyping client software
• Grid markets in telescope time
• More science programmes
• supernova follow-up programme from Sept./Oct.
• collaborative science programme with LANL
• More telescopes, more of the time

25
Future deployment

• Moving out of beta mode at the JAC
• will now be provided as a common user facility
• probably available from next semester
• still to decide which types of observation
• Handshake agreements for interoperability/deployme
  nt
• RAPTOR/TALONS
• MONET
• NOAO
• ING
• Others?

26
Summary

• Event driven observing
• IVOA standard VOEvent messaging
• HTN standard RTML messaging
• eSTAR represents a turn-key system
• Real science
• GRB and exo-planet observing programmes
• See www.estar.org.uk/ for more information