Information Technology Reference
In-Depth Information
Machines (JVMs), which may not suitable for
very resource limited devices. Special design con-
siderations are needed for those wireless devices
that have limited network bandwidth and power.
Service Status Inquiry. A client may be inter-
ested in services' events or status changes. One way
of knowing about them is by polling the service.
Another way, known as service event notification,
is by registering with the service and the service
will notify clients who have shown interest. Most
protocols implement service event notifications.
If events are generated very frequently or a ser-
vice status changes very fast, it is better to use
service polling.
It is even better to have agents do event filtering
and aggregation. Jini provides several such meth-
ods. Services send events to agents and let agents
make sure all the events are delivered to clients;
an agent may act as a sink for events, which will
be filtered, aggregated, and then sent to clients;
or an agent may also resemble a mailbox to filter
events over time. Although clients and services
benefit from event filtering and aggregation, some
resources (computers in networks) need to provide
the functions and handle the events.
Many service discovery protocols use distrib-
uted directories, which store services' information
within their own domains. Service information is
distributed among directories. A directory failure
only affects part of the system. With less infor-
mation in each directory, service lookup within a
directory is more efficient. On the other hand, a
service lookup may go through several directories.
In contrast, a service lookup in a centralized direc-
tory only goes to one directory with less network
communication overhead and latency.
Storage of Service Information. For each
service, the service information may be a single
copy, multiple copies, or fully replicated in di-
rectories. Many protocols have a single copy of
services in its domain. A directory failure will
affect the domain for which it is responsible. In
Jini and SLP service discovery environments,
multiple directories may coexist. Therefore, mul-
tiple copies of service information may exist. It
is more reliable with multiple copies of service
information in several directories, but the greater
the number of directories, the greater the overhead.
INS implements fully replicated copies within a
sub domain. The advantage of fully replicated
directories is that a service search only goes to
the directory to which a client is attached. Mul-
tiple copies or fully replicated copies of service
information should be consistent in directories.
Otherwise, querying different directories may
result in different service information and may
cause problems.
Directory Structure. In a flat directory
structure, directories maintain peer-to-peer rela-
tionships and are equally important. In one type
of flat directory structure, directories connect to
each other and exchange information. In INS, for
instance, directories have a mesh structure and
exchange information with other directories, so
that all service information is available locally
and the service search is very efficient. These
information exchanges generate much commu-
nication traffic, and therefore it is not scalable.
INS/Twine based on peer-to-peer technology is
Directory Designs
Directories cache service information and answer
clients' lookup requests. Thus, the overhead of
handling unrelated requests for services and the
communication between clients and unrelated
services are removed. More importantly, this fa-
cilitates large-scale service discovery. Directory
architectures, service information cache strate-
gies, and hierarchies are different depending on
the environments.
Centralized vs. Distributed Directories.
A centralized directory stores all the services'
information in a central location. The directory
is likely to be a bottleneck and the single point of
failure, which causes the whole system's failure.
In large service discovery domains, it is inefficient
to go through a centralized directory all the time.
Search WWH ::




Custom Search