Barbara Guidi’s seminar
By stefania | October 4, 2011
Speaker: Barbara Guidi
Date: 2011 October, 6
Time: 11:00 a.m.
Duration: 60′
Room: C-40
Title: Reti P2P
Abstract:
Di particolare interesse sono le reti P2P geografichefiche, nelle quali un nodo è caratterizzato da delle coordinate spaziali de_nite su di un piano d-dimensionale.
Questi overlay network necessitano di essere strutturati verso una topologia che permetta di sfruttare le informazioni spaziali associate ai nodi. Una particolare struttura molto utilizzata per modellare queste particolari soluzioni è la triangolazione di Delaunay. Le triangolazioni di Delaunay sono triangolazioni ben note in geometria computazionale e presentano particolari proprietà che risultano utili se applicate in reti geografiche.
Il seminario presenterà un algoritmo innovativo per la costruzione di un overlay network basato sulla triangolazione di Delaunay, sfruttando l’informazione ottenuta in seguito ad una ricerca epidemica all’interno dello overlay network. L’utilizzo del protocollo gossip permette di venire a conoscenza entro un numero limitato di cicli dei nodi presenti nella rete. Tale informazione viene analizzata ed elaborata dall’algortimo proposto per la creazione e la gestione dello overlay basato sulla
triangolazione di Delaunay. Attraverso l’uso della tecnica gossip è garantita la convergenza di un overlay network verso una triangolazione di Delaunay. La particolarità del protocollo proposto è la possibilità di convergere verso una triangolazione di Delaunay sfruttando soltanto l’informazione sui nodi presenti nella rete contenuta all’interno della vista di ogni nodo.
Topics: Uncategorized | No Comments »
Last Papers
By admin | March 8, 2011
HPC Lab Upcoming Papers
- D. Broccolo, L. Marcon, F. M. Nardini, F. Silvestri, R. Perego. Generating Suggestions for Queries in the Long Tail with an Inverted Index. IP&M. To Appear, 2011.
- C. Macdonald, I. Ounis, N. Tonellotto. Upper Bound Approximations for Dynamic Pruning. ACM Transactions on Information Systems (ACM TOIS), to appear, 2011.
- G. De Francisci Morales, A. Gionis, M. Sozio. Social Content Matching in MapReduce. 37th International Conference on Very Large Data Bases (VLDB 2011), Aug 29-Sep 3, Seattle, US
- G. Capannini, F.M. Nardini, R. Perego, F. Silvestri. Efficient Diversification of Web Search Results, 37th International Conference on Very Large Data Bases (VLDB 2011), Aug 29-Sep 3, Seattle, US.
- A. Orlandi, R. Venturini. Space-efficient Substring Occurrence Estimation. 30th ACM SIGMOD-SIGACT-SIGART Symposium on Principles of Database Systems (PODS), 2011.
- M. Boley, C. Lucchese, D. Paurat and T. Gärtner. Direct Local Pattern Sampling by Efficient Two-Step Random Procedures. 17th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2011).
- Gabriele Capannini, Fabrizio Silvestri, Ranieri Baraglia. Sorting on GPUs for large scale datasets: A thorough comparison. Information Processing & Management, In Press, Available online from 8 January 2011.
- C. Lucchese, S. Orlando, R. Perego, F. Rabitti. Similarity Caching in Large-Scale Image Retrieval. Information Processing & Management, In Press, Available online from 8 January 2011.
- R. Baraglia, G. Capannini, D. Laforenza, M. Pasquali, L. Ricci. A multi-level scheduler for batch jobs on grids, The Journal of Supercomputing, In Press, Available online from 22 February 2011.
- G. Capannini, F.M. Nardini, R. Perego, F. Silvestri. Efficient Diversification of Search Results using Query Logs (poster), WWW’11, 28th March – 1st April, Hyderabad, India.
- F. Bonchi, R. Perego, F. Silvestri, H. Vahabi, R. Venturini. Recommendations for the Long Tail by Term-Query Graph (poster), WWW’11, 28th March – 1st April, Hyderabad, India.
- C. Lucchese, S. Orlando, R. Perego, F. Silvestri, G. Tolomei. Identifying Task-based Sessions in Search Engine Query Logs (best paper runner up). ACM WSDM, Hong Kong, February 9-12, 2011.
- D. Ceccarelli, C. Lucchese, S. Orlando, R. Perego, F. Silvestri. Caching query-biased snippets for efficient retrieval (full paper). EDBT 2011, Upsala. Sweden, March 22-24, 2011.
- R. Perego, F. Silvestri, N. Tonellotto. Representing Document Lengths with Identifiers (poster). ECIR, Dublin, Ireland, April 18-21, 2011.
Topics: Uncategorized | No Comments »
Montresor’s seminar
By stefania | February 8, 2011
Speaker: Prof. Alberto Montresor
Date: 2011, February, 21
Time: 12:00 p.m.
Duration: 40/45′
Room: Aula Faedo (C-29)
Title: Cloudy Weather for P2P, with a Chance of Gossip
Abstract:
Peer-to-peer (P2P) and cloud computing, two of the Internet trends of the last decade, hold similar promises: the (virtually) infinite availability of computing and storage resources. But there are important differences: the cloud provides highly-available resources, but at a cost; P2P resources are for free, but their availability is shaky. Several academic and commercial projects have explored the possibility of mixing the two, creating a large number of \emph{peer-assisted} applications, particularly in the field of content distribution, where the cloud provides a highly-available and persistent service, while P2P resources are exploited for free whenever possible to reduce the economic cost. While executing active servers on elastic computing facilities like Amazon EC2 and pairing them with user-provided peers is definitely one way to go, this talk proposes a novel approach that further reduces the economic cost. Here, a passive storage service like Amazon S3 is exploited not only to distribute content to clients, but also to build and manage the P2P network linking them. An effort is made to guarantee that the read/write load imposed on the storage remains constant, regardless of the number of peers/clients. These two choices allows us to keep the monetary cost of the cloud always under control, in the presence of just one peer or with a million of them. We show the feasibility of our approach by discussing two cases studies for content distribution.
Topics: Uncategorized | No Comments »
Ceccarelli’s seminar
By stefania | December 10, 2010
Speaker: Diego Ceccarelli
Date: December 13
Duration: 55′
Time: 10:00 a.m.
Room: C-29
Title: The Web of Objects
Abstract:
Nowadays, there are already hundreds of millions of documents embedding metadata (e.g. RDF, RDFa, microformats …), in addiction traditional web pages usually contain structured data organized in lists or tables. (e.g., infoboxes in Wikipedia).
In state of the art the web is modeled as a collection of hyper-linked pages (each containing a bag of words) i.e., a document-centric model. In this talk we present web of objects: an entity-centric
model, where entities (e.g., restaurants, universities, persons) are connected through semantic relations.
We discuss several research challenges and propose some ideas.
Topics: Uncategorized | No Comments »
Righetti’s seminar
By stefania | November 22, 2010
Speaker: Giacomo Righetti
Date: December 6
Duration: 55′
Time: 11:00 a.m.
Room: C-29
Title: Intermediate Code Execution on GPU: the VMs meet the graphical
processors
Abstract:
Today, language-oriented virtual machines such as the JVM and the .NET CLR are successfully used in many areas to provide several advantages, most notably greater portability and software reuse. However, these systems provide a bytecode which targets a purely sequential computational model, and is not directly mapped on recent GPU architectures, as they exploit a computational model based on the data parallel paradigm. A mapping of conventional bytecode onto the GPU instruction set, if realized, would allow a much easier and efficient use of the large raw computational power the GPU can provide.
In this thesis we have realized a meta-program that can generate the GPU version of a .NET computation. This meta-program performs bytecode analisys and creates a semantically equivalent program. The approach we followed relies on the transformation of compiled programs in binary format (Assembly), exploiting high-level semantic annotations, which express among other things the user types, and are preserved in the Assembly by the CLR compilation process. The programmer thus develops only the sequential version, marking at high level with dedicated object types and methods the potentially data-parallel code sections, but delaying the exact decision of what and how to parallelize at runtime. The thesis work has led to the development of the PBricks library, which realizes a mapping between the CLR assemblies and the GPUs executable code. We identified a precise memory model and a translator from the CLR bytecode to the ATI GPU intermediate language. They are sufficiently general to be extended to other virtual execution environments with introspection capabilities, like the Java JVM, and to target execution on any modern GPU, e.g. on Nvidia GPUs.
Topics: Uncategorized | No Comments »
Franco Maria Nardini has given a talk at RIAO 2010
By claudio | April 30, 2010
Franco Maria Nardini, one of our students has just given a talk at RIAO 2010 in Paris on the paper entitled: “The Effects of Time on Query Flow Graph-based Models for Query Suggestion”.
Here‘s a brief clip of his talk.
Topics: Uncategorized | No Comments »
Venturini’s seminar
By stefania | April 12, 2010
Speaker: Rossano Venturini
Date: ?
Duration: 55′
Time: 10:00 a.m.
Room: C-40
Title: Efficient coding and fast decoding of integer lists.
Abstract:
In this talk I will present a novel (class of) encoder to compress lists of integers.
Roughly speaking, the encoder works by partitioning the list of integers into blocks and
by compressing each of them separately by resorting to very simple encoders.
I will show a way to find the optimal partition (namely, the one that maximizes the
achieved compression) and a experimental comparison between the new method and
some of the best known compressors.
Topics: Uncategorized | No Comments »
Grasso’s seminar
By stefania | April 7, 2010
Speaker: Ivan Grasso
Date: 2010, April 8
Duration: 55′
Time: 10:00 a.m.
Room: C-40
Title: (Italian) “Implementazione ed Analisi di un IR Query Processor su Architettura Cell/B.E”
Abstract: (Italian)
“Il web nel corso degli anni ha avuto una continua crescita sia in termini di informazioni pubblicate, sia in termini di utenti. Per gli utilizzatori della rete quindi diventa sempre più difficile orientarsi e per questo motivo i Motori di Ricerca sono sempre più utilizzati.
Ogni giorno infatti gestiscono milioni di richieste sottomesse dagli utenti. In questo scenario, le performance sono di primaria importanza e per gestire l’enorme quantità di dati si utilizzano sistemi massivamente paralleli costituiti da migliaia di macchine o, per usare un termine più moderno sistemi cloud.
In sistemi di tale grandezza ha acquisito sempre più importanza anche il fattore di consumo, portando negli ultimi tempi la ricerca, scientifica ed industriale, a investigare meccanismi di query processing più efficienti e meno dispendiosi. Una possibile soluzione è stata cercata attraverso l’utilizzo di GPU (graphical processing unit) dimostrando la possibilità di aumentare significativamente il numero di query elaborate al secondo [DHYS09].
In questa tesi ci proponiamo di implementare un IR Query Processor sull’architettura Cell/B.E., che affianca ad un generico processore una serie di acceleratori asincroni con ampie possibilità di calcolo [CRDI07]. La tesi si concentrerà su tale architettura, sottolineandone pregi e difetti durante l’implementazione e la successiva analisi del modello software realizzato”
Topics: Uncategorized | No Comments »
Montresor’s seminar
By stefania | February 22, 2010
Speaker: Alberto Montresor
Date: 2010, February 24
Duration: 55′
Time: 02:30 p.m.
Room: C-29
Title: Gossip in Peer-to-Peer: Beyond dissemination
Abstract:
The gossip paradigm made its first appearance in distributed systems in 1987, when it was applied to disseminate updates in replicated databases. Two decades later, gossip-based protocols have gone far beyond dissemination, solving a large and diverse collection of problems. We believe that the story is not over: while gossip is not the panacea for distributed systems, there are still virgin research areas where it could be profitably exploited. In this talk, we briefly discuss a gossip-based “construction set” for distributed systems and we illustrate how large-scale distributed computing could benefit by the application of its building blocks.
Topics: Uncategorized | No Comments »
Capannini’s seminar
By stefania | February 8, 2010
Speaker: Gabriele Capannini
Date: 2010, February, 26
Duration: 55′
Time: 02:30 p.m.
Room: C-29 (Faedo room)
Title:K-Model: A New Computational Model for Stream Processors
Abstract:
We introduce k-model, a computational model to properly evaluate algorithms designed for graphic processors, and other architectures adhering to the stream programming model. We address the lack of one formal complexity model that properly accounts for memory contention, address coalescing in memory accesses, or the serial control of the instruction flows. We study the impact of k-model rules on algorithm design. We devise a coalesced and low contention data access technique for Batcher’s networks, and we evaluate the effectiveness of this technique within our k-model. To evaluate the benefits in using k-model in evaluating solutions for streaming architectures, we compare the complexity of a sorting network built using our technique, and quicksort. Although in theory quicksort is more efficient than bitonic sort, empirically, our bitonic sorting network has been shown to be faster than the state-of-the-art implementation of quicksort on graphics processing units (GPUs). Using our k-model we are able to prove the reason why on GPU architectures this is not true anymore. As a side result, our technique to perform a Batcher’s network on GPUs improves the performance of the fastest comparison-based solution for integers sorting.
Topics: Uncategorized | No Comments »
« Previous Entries