Modeling P2P Grid Information Services with Colored Petri Nets - Grid Computing: Infrastructure, Service, and Applications - page 327

13.1
The past few years have witnessed a rapid development of grid comput-
ing systems and their applications. Information services play a crucial role
in grid computing environments to facilitate the discovery of resources
and services. Existing work on information services, notably the Monitor-
ing and Discovery System (MDS4) [1] of the Globus middleware [2], has
advanced the research in this i eld. MDS4 adopts a hierarchical tree struc-
ture to distribute its monitoring load across a virtual organization (VO), in
which every node is running an index service monitoring its resources
and pushing this information up to a master index server. Thus a VO has
a central point for registering the state information of computing resources
involved in the VO.
However, due to the dynamic nature of the grid, keeping an accurate
and up-to-date source of state information poses many challenges in
terms of scalability, resilience, and performance. In the case of MDS4, its
hierarchical structure and centralized management has an inherent delay
associated with it, limiting its scalability. The work presented in [3] shows
that MDS4 has a slow response even with a load of 100 nodes. MDS4 also
lacks a mechanism to deal with index service node failures, which could
break the information service network into isolated subnets.
Many factors exist that can inl uence the performance of an information
service system. For example, slow network connections may produce trafi c
bottlenecks in communication between nodes, and frequently changing
resources require more updates to be sent out to the information service
network. The current workload being undertaken by key index nodes could
also slow down the updates of information, while the type of index nodes
in question could also be varied in terms of computing capabilities. These
are just some of the obstacles that could hinder the accuracy of information
and consequently limit the efi ciency of large-scaled grid environments.
Peer-to-peer (P2P) systems have shown enhanced scalability and fault
tolerance in i le sharing under a highly dynamic computing environment.
Distributed hash table (DHT)-based second-generation P2P systems such
as Chord [4], Pastry [5], and CAN [6] use DHTs to speed up the process of
message lookup. A message lookup can be completed within a guaran-
teed number of hops; for example, both Pastry and Chord have a maxi-
mum of O log( n ) hops, with CAN differing with O(d n 1/d) hops. However,
the common weakness exhibited by using a DHT is the lack of ability to
deal with churn situations in which nodes join or leave at high rates [7].
One side-effect of the churn situation is that it could lead to a high l uctua-
tion in both communication and computational cost, causing extra heavy
overhead to a P2P network.
Building on grid and P2P technologies, P-Grid [8] bases its network
on a binary tree with each leaf of the tree representing a computer node.
Data are distributed using hash keys that dictate where the data will be

Introduction