With the constant increase in the number of interconnected devices in today networks, and the high demand of adaptiveness, more and more computations can be designed according to self-organisation principles. In this context, a key building block for large-scale system coordination, called gradient, is used to estimate distances in a fully-distributed way: it is the basis for a vast variety of higher level patterns including information broadcast, events forecasting, distributed sensing, and so on. However, computing gradients is very problematic in mobile environments: the fastest self-healing gradient conceived so far (called BIS) achieves a reaction speed proportional to the single-path speed of information in the network. In this paper we introduce a new gradient algorithm, SVD (Stale Values Detection) gradient, which uses broadcasts to reach a reaction speed that is equal to the multi-path speed of information, namely, the fastest speed possibly achievable by network algorithms. We then combine SVD with other blocks (metric correction, smooth filtering, BIS gradient, information damping) proposing a composed block called ULT(imate) gradient. We evaluate the resulting algorithm and compare it with other approaches, showing it scores best both on accuracy and smoothness while keeping communication cost under control.

Compositional Blocks for Optimal Self-Healing Gradients

Audrito, Giorgio;Damiani, Ferruccio;
2017-01-01

Abstract

With the constant increase in the number of interconnected devices in today networks, and the high demand of adaptiveness, more and more computations can be designed according to self-organisation principles. In this context, a key building block for large-scale system coordination, called gradient, is used to estimate distances in a fully-distributed way: it is the basis for a vast variety of higher level patterns including information broadcast, events forecasting, distributed sensing, and so on. However, computing gradients is very problematic in mobile environments: the fastest self-healing gradient conceived so far (called BIS) achieves a reaction speed proportional to the single-path speed of information in the network. In this paper we introduce a new gradient algorithm, SVD (Stale Values Detection) gradient, which uses broadcasts to reach a reaction speed that is equal to the multi-path speed of information, namely, the fastest speed possibly achievable by network algorithms. We then combine SVD with other blocks (metric correction, smooth filtering, BIS gradient, information damping) proposing a composed block called ULT(imate) gradient. We evaluate the resulting algorithm and compare it with other approaches, showing it scores best both on accuracy and smoothness while keeping communication cost under control.
2017
11th IEEE International Conference on Self-Adaptive and Self-Organizing Systems, SASO 2017
University of Arizona, usa
18 September 2017 through 22 September 2017
Proceedings - 11th IEEE International Conference on Self-Adaptive and Self-Organizing Systems, SASO 2017
Institute of Electrical and Electronics Engineers Inc.
91
100
9781509065554
http://ieeexplore.ieee.org/document/8064033/
adaptive algorithm; aggregate programming; computational field; gradient; Artificial Intelligence; Control and Optimization
Audrito, Giorgio; Casadei, Roberto; Damiani, Ferruccio; Viroli, Mirko
File in questo prodotto:
File Dimensione Formato  
IEEE-SASO-2017-Audrito-et-al.pdf

Accesso riservato

Descrizione: Articolo principale (conferenza)
Tipo di file: PDF EDITORIALE
Dimensione 408.7 kB
Formato Adobe PDF
408.7 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
IIEEE-SASO-2017-Audrito-et-al-OPEN.pdf

Accesso aperto

Descrizione: Articolo principale (conferenza)
Tipo di file: POSTPRINT (VERSIONE FINALE DELL’AUTORE)
Dimensione 816.42 kB
Formato Adobe PDF
816.42 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2318/1655254
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 46
  • ???jsp.display-item.citation.isi??? 35
social impact