collaborative task execution in volunteer clouds (or how to choose a sub-reviewer)

Post on 02-Jul-2015

594 Views

Category:

Technology

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

My talk at the 2nd General Meeting of the CINA project, Bologna, 18-20 Feb 2014. The increasing diffusion of cloud technologies offers new opportunities for distributed and collaborative computing. Volunteer clouds are a prominent example, where participants join and leave the platform and collaborate by sharing computational resources. The high complexity, dynamism and unpredictability of such scenarios call for decentralized self-* approaches. We present in this paper a framework for the design and evaluation of self-adaptive collaborative task execution strategies in volunteer clouds. As a byproduct, we propose a novel strategy based on the Ant Colony Optimization paradigm, that we validate through simulation-based statistical analysis over Google workload data.

TRANSCRIPT

Collaborative Task Execution

In Volunteer Clouds

-- Michele Amoretti, PARMA

-- Alberto Lluch Lafuente, IMT

-- Stefano Sebastio, IMT

2nd General Meeting, Bologna, 18-20 Feb 2014

Collaborative Task Execution in

Volunteer Clouds

-- Michele Amoreti, PARMA

-- Alberto Lluch Lafuente, IMT

-- Stefano Sebastio, IMT

Collaborative Task Execution in

Volunteer Clouds

-- Michele Amoreti, PARMA

-- Alberto Lluch Lafuente, IMT

-- Stefano Sebastio, IMT

Collaborative Task Execution in

Volunteer Clouds

-- Michele Amoreti, PARMA

-- Alberto Lluch Lafuente, IMT

-- Stefano Sebastio, IMT

How to choose

a sub-reviewer

How to choose

a sub-reviewer

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

X X X XX X X XX

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

X X X XX X X XX

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

X X X XX X X XX

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

X X X XX X X XX

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

X X X XX X X XX

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

X X X XX X X XX

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

choose reviewers

(almost) randomly

SciFi Community

SciFi Community1 Unstructured network

SciFi Community1

2

Unstructured network

All members generate review tasks

SciFi Community1

2

3

Unstructured network

All members generate review tasks

All members perform reviews

SciFi Community1

2

3

4

Unstructured network

All members generate review tasks

All members perform reviews

Review requests may be forwarded

SciFi Community1

2

3

4

5

Unstructured network

All members generate review tasks

All members perform reviews

Review requests may be forwarded

All members apply the same algorithm

SciFi Reviewers

1

SciFi Reviewers

No rescheduling, no priorities.

1

2

SciFi Reviewers

No rescheduling, no priorities.

Accept request iff CoS met.

1

2

3

SciFi Reviewers

No rescheduling, no priorities.

Accept request iff CoS met.

No delays.

1

2

3

SciFi Reviewers

No rescheduling, no priorities.

Accept request iff CoS met.

No delays.

Reply/Forward requests immediately.4

1

2

3

SciFi Reviewers

No rescheduling, no priorities.

Accept request iff CoS met.

No delays.

Reply/Forward requests immediately.

Disclose confidence on research topics.

4

5

ALGORITHM 1:

RANDOM

Algorithm 1: Random

Each outgoing arc has the same probability of being chosen during request propagation.

Algorithm 1: Random

Each outgoing arc has the same probability of being chosen during request propagation.

Algorithm 1: Random

Each outgoing arc has the same probability of being chosen during request propagation.

Algorithm 1: Random

ALGORITHM 2:

Greedy ORACLE

The GreedyORACLE

The oracle provides the sub-reviewer who will finish earlier.

The GreedyORACLE

The oracle provides the sub-reviewer who will finish earlier.

The GreedyORACLE

The oracle provides the sub-reviewer who will finish earlier.

ALGORITHM 3:

FEEDBACK BASED

Probabilistic routingArcs are labelled with rates to be used in probabilistic choices.

11

1

1

1+1 1

Feedback-based rates

11

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

Feedback-based rates

11

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2 Can you review?

Feedback-based rates

11

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2 Can you review?

NO

Feedback-based rates

1-11

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2 Can you review?

NO

Feedback-based rates

1-11

1

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

Can you review?

Feedback-based rates

1-11

1

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

Can you review?

Can youreview?

Feedback-based rates

1-11

1

1

1

1+1 1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

Can you review?

Can youreview?

YES

Feedback-based rates

1-11

1

1

1

1+1 1+1

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

Can you review?

Can youreview?

YES

Feedback-based rates

Feedback-based rates

ALGORITHM 4:

con+dence-based

Feedback-based pheromones

FM: 3

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

SE: 1

FM: 4

SE: 5

Arcs labeled with one ratefor each research topic.

Feedback-based pheromones

FM: 3

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

FM: 3SE: 1

SE: 1

FM: 3SE: 1

FM: 4

SE: 5

Arcs labeled with one ratefor each research topic.

Feedback-based pheromones

FM: 3

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

FM: 3SE: 1

SE: 1

FM: 3SE: 1

FM: 0SE: 0

FM: 4

SE: 5

Arcs labeled with one ratefor each research topic.

Feedback-based pheromones

FM: 3

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

FM: 3SE: 1

SE: 1

FM: 3SE: 1

FM: 0SE: 0

FM: 4

SE: 5 SE: 5

FM: 4

Arcs labeled with one ratefor each research topic.

Feedback-based pheromones

FM: 3

Paper XXXDue dd/mm/yyCoS:* FM > 3* SE > 2

FM: 3SE: 1

SE: 1

FM: 3SE: 1

FM: 0SE: 0

FM: 4SE: 5

FM: 4

SE: 5 SE: 5

FM: 4

Arcs labeled with one ratefor each research topic.

Confidence-based Rates

Confidence-based Rates

Confidence-based Rates

choose reviewers

(almost) randomly

1

What's next?

Study the impact of the structure of the overlay network

1

2

What's next?

Study the impact of the structure of the overlay network

Study reputation-based strategies

1

2

3

What's next?

Study the impact of the structure of the overlay network

Study reputation-based strategies

Application to routing of messages in predicate-based communication (cf. SCEL)

Questions?

References

“A Computational Field Framework for Collaborative Task Execution in Volunteer Clouds”, Stefano Sebastio, Michele Amoretti and Alberto Lluch-Lafuente, draft [PDF]

“Reputation-based Cooperation in the Clouds”, Alessandro Celestini, Alberto Lluch Lafuente, Philip Mayer, Stefano Sebastio, and Francesco Tiezzi, draft [PDF]

The science cloud platform. http://svn.pst.ifi.lmu.de/trac/scp/.

P. Mayer et al. The Autonomic Cloud: A Vision of Voluntary, Peer-2-Peer Cloud Computing, 3rd Workshop on Challenges for Achieving Self- Awareness in Autonomic Systems, 2013.

Work-in-progress partially reported in:

See also:

top related