AlgoDEEP Algorithmic challenges for Data-intensivE processing on Emerging computing Platforms 16-17 aprile 20101Kickoff Meeting, Bertinoro.

AlgoDEEP

Algorithmic challenges for Data-intensivE processing on Emerging computing Platforms

16-17 aprile 2010 1Kickoff Meeting, Bertinoro

PRIN 2008: dati Area 9


Progetti cofinanziati: 103

Cofinanziamento medio: 107.700

Cofinanziamento min: 24.810

Cofinanziamento max: 181.608

Com’è andata a noi …


COFIN RICHIESTO

COFIN ASSEGNATO

MESI UOMO

COFIN/MU PERSONE COFIN/PERSONA

232.000 127.027 (54.7%)

442 287 61 2.082

Nota Illustrativa 18/12/1008: “Non potrà essere presa in considerazione […] una riduzione degli obiettivi del progetto […]”perché … ”La riduzione del cofinanziamento, in sede di approvazione del progetto, sarà da porre in relazione esclusivamente con costi ritenuti non congrui dai valutatori”

Costi ritenuti non congrui dai valutatori: nessuno.

Un po’ di nostra storia …


Cofin (Keuro)

TaskTree


CHALLENGE 1Adaptive

Computations

Leader: C. Fantozzi

C1.1: Faulty memories

C1.2: Oblivious multicore computations

CHALLENGE 2Massive Data Analysis

Leader: M. Patrignani

C2.1: Extraction of significant properties and patterns

C2.2: Analysis of dynamic networks

CHALLENGE 3Empowering

Computation on Large-Scale Networks

Leader: F. Grandoni

C3.1: Design and maintainance of large networks

C3.2: Computation and communication in large networks

Kickoff Meeting, Bertinoro16-17 aprile 2010

C1: Adaptive Computations

Adaptivity: ability of reacting to changing conditions.

It is needed, e.g. to ensure correctness

Example: resiliency to hardware faults to attain efficiency

Example: heterogeneous platformsin grid/global/cloud computing

6


C1: Tasks

C1.1: Faulty Memories Design of adaptive algorithms and data structures

that are resilient to memory faults Is ”fault-obliviousness” (i.e., resiliency without

knowing a statistics of faults) possible?

C1.2: Oblivious Multicore Computations How far optimal performance can be attained in

an oblivious fashion w.r.t. parallelism, memory hierarchy, and communication?

Ultimate goal: a notion of ”machine obliviousness”

7


C1: Expected Results

C1.1: Faulty Memories M12: New computational models for faulty memories,

including fault-obliviousness. New resilient (possiblyfault-oblivious) algorithms and data structures

M24: Experimental validation of the resilient algorithmsand data structures developed during the first year

C1.2: Oblivious Multicore Computations M12: Unified computational model that accounts for

network locality, locality of reference, and parallelism,as shaped in multicores

M24: Framework for the specification of machine-oblivious algorithms. Machine-oblivious algorithms for basic primitives

8


C2: Massive Data Analysis

Large datasets are found in biology, marketing, network

analysis, program analysis, etc may be static or dynamic

Challenging tasks extracting significant information (what is

significance?) identifying clusters understanding dynamics

9


C2: Tasks

C2.1: Extraction of Significant Properties & Patterns DYNAMIC PROGRAM ANALYSIS: efficient techniques for

runtime invariant checking and error localization

PATTERN EXTRACTION FROM STATIC DATA: characterization of statistical significance (applied to; frequent itemsets mining); discovery of motifs in genomic sequences

C2.2: Analysis of Dynamic Networks ANALYSIS OF ROUTING AND TRAFFIC INFO: to extract

information about AS graphs dynamics, and clusters

VISUALIZATION TOOLS: dealing with endless streams of data; hybrid visualization through clustering

10



C2.1: Extraction of Significant Properties & Patterns M12: Constraint-based framework for dynamic invariant checking based

on efficient incremental change propagation algorithms. Statistical techniques to guide the mining of frequent itemsets. Extraction of long motifs from genomic sequences.

M24: Efficient external-memory data structures for recording trails of operations. Discovery of motifs with wildcards. Discovery of permutation patterns.

C2.2: Analysis of Dynamic Networks M12: Large repository of routing updates and RIB dumps. Algorithms to

indentify hot-spots, understand routing dynamics, extract structural properties and clusters. Visualization of stream of edges. Visualization of non-planar dense subgraphs and clusters.

M24: Usage of information gathered to optimize and reduce instability of BGP and detecting causes of instability in real-time. Implementation and experimental validation of clustering algorithms. Tools to visualize clustered graphs and stream of edges. algorithms for the visualization of non-planar networks.

11


C3: Computation on Large Networks

Analysis: structural properties of real-world networks

Design: networks with complex constraints

Maintenance: algorithmic and visualization tools for network monitoring and maintenance

Use: algorithms and data structures for efficient computation and communication

12


C3: Tasks

C3.1: Design and Maintenance NETWORK DESIGN: design and analysis in complex

scenarios

NETWORK MAINTAINANCE: Monitoring and maintenance of large-scale heterogeneous networks

C3.2:Computation and Communication EMPOWERING COMPUTATION: Space/network-efficient

algorithms for cloud computing

EMPOWERING COMMUNICATION: Efficient communication primitives for routing/broadcasting

13

AlgoDEEP Kickoff MeetingApril 16-17, 2010


C3.1: Design and Maintenance M12: Multi-objective network design. Low-stretch routing with

limited node memory. Analysis of static wireless networks. Techniques for real-time monitoring of traffic. Maintenance of tree spanners under network failures.

M24: Stochastic network design and experimental validation of developed algorithms. Near-optimal bandwidth usage with limited node memory. Analysis of mobile wireless networks. Implementation of real-time monitoring tools. Distributed maintenance of robust spanners.

C3.2:Computation and Communication M12: Service-oriented computing with outsourcing of computation.

Models for distributed/streaming computation, with applications to cloud computing. Labeling schemes for routing/broadcasting. Heuristics and visualization tools for stability check.

M24: Development of synopses. Primitives for processing massive text collections. Experimental validation of labeling schemes. Testing of heuristics for instability detection.



Units Involved

C1.1 C1.2 C2.1 C2.2 C3.1 C3.2

Padova * * *Pisa + * * *Roma 1 * + * * + *Roma 2 * + + * *Roma 3 * * *

The symbol * denotes direct involvement The symbol + denotes a mere interest

15

Da proposal:

Kickoff Meeting, Bertinoro16-17 aprile 2010 16

1. Kickoff meeting

2. Meeting fine I anno ?

3. Meeting finale

4. Sito web

(verona.dei.unipd.it/~prin08)


Honors

17

• F. Grandoni et al.:

Best Paper Award STOC2010

• F. Silvestri et al.:

Best Paper Award IPDPS 2010

APPLAUSO


Pianificazione

18

• Sito web (verona.dei.unipd.it/~prin08)

• lucidi kickoff

• Rideterminazione dei Costi• Scadenza ufficiale 3/5/2010 ore 17

• Scadenza nostra 28/4/2010• “ripartizione senza alcun vincolo”. In fase di esecuzione, eventuali

rimodulazioni delle spese senza richiesta di autorizzazione, non devono superare il 20% di quanto deciso in fase di rideterminazione

• Valutazione finale: occhio agli expected results

• Workshop tematici

• Varie ed eventuali


Pianificazione

19


KICKOFF Meeting

20

AlgoDEEP Algorithmic challenges for Data-intensivE processing on Emerging computing Platforms 16-17 aprile 20101Kickoff Meeting, Bertinoro.

Documents

kickoff meeting

large networks analysis

bertinoro slide

networks design

patterns c2

machineoblivious algorithms

massive data analysis

network analysis