UNIVERSAL IDENTIFICATION OF LOCATION SERVICE

Abstract

We describe a decentralized, adaptive mechanism for replica location in wide-area distributed systems. Unlike traditional, hierarchical (e.g, DNS) and more recent (e.g.,CAN,  Chord,  Gnutella)  distributed  search  and indexing schemes,  nodes  in  our  location mechanism  do  not  route queries,  instead,  they  organize  into  an  overlay  network and distribute  location  information. We  contend  that  this approach  works  well  in  environments  where  replica location queries are prevalent but the dynamic component of  the  system  (e.g.,  node  and  network  failures,  replica add/delete operations) cannot be neglected.  We  argue  that  a  replica  location  mechanism  that combines probabilistic representations of replica location information  with  soft-state  protocols  and  a  flat  overlay network  of  nodes  brings  important  benefits:  genuine decentralization,  low  query  latency,  and  flexibility  to introduce adaptive communication schedules.  We support these claims in two ways. First, we provide a  rough  resource  consumption  evaluation: we  show  that, for  environments  similar  to  those  encountered  in  large scientific data analysis projects, generated network traffic is  limited  and,  more  importantly,  is  comparable  to  the traffic generated by a request routing scheme. Second, we provide  encouraging performance  data  from  a prototype implementation.