Ci-dessous, les différences entre deux révisions de la page.
Les deux révisions précédentes Révision précédente Prochaine révision | Révision précédente Prochaine révision Les deux révisions suivantes | ||
site:recherche:axes_de_recherche [15/12/2017 11:09] amann |
site:recherche:axes_de_recherche [15/12/2017 22:01] amann |
||
---|---|---|---|
Ligne 4: | Ligne 4: | ||
====== Axes de recherche ====== | ====== Axes de recherche ====== | ||
- | <WRAP tabs> | + | {{page>site:menu}} |
- | * [[site:|Accueil]] | + | |
- | * [[site:recherche:start|Recherche]] | + | |
- | * [[site:enseignement:start|Enseignement]] | + | |
- | * [[site:recherche:axes_de_recherche|Axes de Recherche]] | + | |
- | </WRAP> | + | |
===== Data streams and continuous queries ===== | ===== Data streams and continuous queries ===== | ||
- | The web produces continuous streams of text items | + | The web produces continuous streams of text items published as RSS news, tweets, blog messages etc. Users can subscribe to these streams be defining queries which continuously filter and rank the most recent information items. A major challenge is then to efficiently process millions of such subscription queries over high rate input streams. In the context of the ROSES ANR project ROSES (2008-2012) on RSS feed aggregation and filtering we worked in on multi-query optimisation (PhD J. Creus), on efficient refresh strategies for dynamic RSS feeds (PhD of R. Horincar in collaboration with the MLIA team), and on continuous top-k query processing (PhD of N. Vouzoukidou in collaboration with ICS-Forth, Crete). |
- | published as RSS news, tweets, blog messages etc. Users can subscribe to these streams be defining queries | + | |
- | which continuously filter and rank the most recent information items. A major challenge is then to efficiently | + | |
- | process millions of such subscription queries over high rate input streams. In the context of the ROSES ANR | + | |
- | project ROSES (2008-2012) on RSS feed aggregation and filtering we worked in on multi-query optimisa- | + | |
- | tion (PhD J. Creus), on efficient refresh strategies for dynamic RSS feeds (PhD of R. Horincar in collaboration | + | |
- | with the MLIA team), and on continuous top-k query processing (PhD of N. Vouzoukidou in collaboration with | + | |
- | ICS-Forth, Crete). | + | |
===== Web archive indexing and maintenance ===== | ===== Web archive indexing and maintenance ===== | ||
Ligne 35: | Ligne 24: | ||
===== Workload-aware data replication ===== | ===== Workload-aware data replication ===== | ||
- | Distributed transactions in large data clusters | + | Distributed transactions in large data clusters |
generate a high control and synchronization overhead which is a major obstacle for achieving scalability. To | generate a high control and synchronization overhead which is a major obstacle for achieving scalability. To | ||
reduce this overhead, we focus on user-centric applications where (1) the data fragment attached to each user | reduce this overhead, we focus on user-centric applications where (1) the data fragment attached to each user | ||
- | definesthebasicaccessunit,(2)transactionsmostlyaccessthedataoftwousers(messageexchange)and(3)the | + | defines the basic access unit, (2) transactions mostly access the data of two users (message exchange) and (3) the |
access frequency (popularity) is biased and fluctuates over time. To achieve optimal performance, we propose | access frequency (popularity) is biased and fluctuates over time. To achieve optimal performance, we propose | ||
to move user data to a single node where the transaction can be executed locally. Then, under the assumption | to move user data to a single node where the transaction can be executed locally. Then, under the assumption |