Efficient Data Grids should provide users with services for optimised access to data. In the design of data access services is therefore extremely relevant to evaluate and compare adaptable data selection algorithms, especially when location and number of file copies are expected to change dynamically over time. In this paper we use Petri Nets to formally evaluate candidate protocols to be used by data access services for selection of Grid-wide replicated data files that need to be analysed by Grid jobs. In particular, we compare a centralised Replica Catalogue-based protocol with a Peer-2-Peer protocol, where there is no centralised knowledge about data location. Our results show that the Peer-to-Peer protocol is more scalable than the centralised one and outperforms it when used in Data Grids having more than few sites
Analysis of Replica Selection Protocols for Grid Data Access Services
Zini, Floriano
2004-01-01
Abstract
Efficient Data Grids should provide users with services for optimised access to data. In the design of data access services is therefore extremely relevant to evaluate and compare adaptable data selection algorithms, especially when location and number of file copies are expected to change dynamically over time. In this paper we use Petri Nets to formally evaluate candidate protocols to be used by data access services for selection of Grid-wide replicated data files that need to be analysed by Grid jobs. In particular, we compare a centralised Replica Catalogue-based protocol with a Peer-2-Peer protocol, where there is no centralised knowledge about data location. Our results show that the Peer-to-Peer protocol is more scalable than the centralised one and outperforms it when used in Data Grids having more than few sitesI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
