Grid Resource Allocation - Grid Computing: Infrastructure, Service, and Applications - page 178

BitTorrent
Napster

Deterministic

Unstructured

Gnutella

Non-
deterministic

Kazaa

P2P
network

Pastry
Tapestry
Chord
Koorde
CAN

DHT

Structured

Non-DHT Mercury

FIGURE 7.5 Peer-to-peer network taxonomy.

node sends back a bid that contains the charge for the job to the
original node; otherwise, it forwards the request to other nodes
based on the above propagation approaches.

3. The consumer collects all the bids and chooses the one charging
the least, and then sends the job offer as well as job parameters to
the selected node.

4. The selected node inserts the job into its job queue once it receives
the offer and sends back the results.

A grid emulator was proposed to evaluate the performance of a
combined environment: a large-scale grid with P2P systems' dynamism
and intermittent participation properties. Results give a quantitative
measure of the sharing environment and can be found in [22].
How to discover and allocate the resources in this combined resource-
sharing environment is not trivial. Some issues, such as integrating task
scheduling into grids and efi cient request propagation approaches, are
currently active research topics.

7. 4 . 5 Advanced Reservation

There are two scenarios for resource allocation. One is for metacomputing
applications such as TeraGyroid [27] and SPICE [28]. This kind of applica-
tion requires all the resources to be available at the same time. Another
scenario is for workl ow applications such as Pegasus [29]. In this scenario,
resources need to be scheduled at different times to guarantee task
turnaround. Each task cannot be executed unless all its dependent tasks
are completed. Resource reservation is one solution to these allocation
concerns. It allows users to request resources in a proper order and guar-
antees access to sufi cient resources for their applications.