From its inception a principal goal of the Next Generation Internet (NGI) h
as been to find a way to provide reliable, scalable, cost effective, and de
ployable delivery of data with quality of service (QoS) guarantees as a fou
ndation for innovative NGI applications. But while schemes to provide relia
ble end-to-end QoS are being actively pursued on a number of fronts, the we
ll known problems with the scalability, cost, and deployability of end-to-e
nd QoS continue to obstruct progress toward achieving this end. Our researc
h focuses on the nature and potential value of an approach to providing QoS
that builds on a strategy for allowing NGI applications to dynamically man
age remote storage resources in order to stage data locally for later deliv
ery. We call this strategy logistical Quality of Service (logistical QoS).
The concept of logistical QoS is a generalization of the typical end-to-end
model for reserving QoS, permitting much more flexible use of buffering of
messages in order to achieve QoS delivery without difficult end-to-end req
uirements. Whenever data is available to be sent well before it needs to be
received, it can be staged, i.e. moved in advance to a location close to t
he receiver for later delivery. Isolating the act of buffering data as a di
stinct operation, independent of delivery to the receiver, opens up a new d
imension of freedom in the management of communication and storage resource
s that can offer NGI application developers a wide variety of new opportuni
ties to innovate.
Our project focuses on the development of logistical QoS as enabling networ
k functionality for application-driven staging and scheduling of distribute
d computation on NGI. It is divided into two parts: (1) research on the bas
ic network functionality that is required to support logistical QoS that is
reliable, scalable, cost effective, and easy to use; and (2) research that
investigates the integration of logistical QoS and the basic network techn
ology that underlies it with the scheduling of distributed computations usi
ng NetSolve and the Network Weather Service.
Our work on logistical QoS focuses on the Internet Backplane, as providing
a mechanism for managing remote storage resources, and the Internet Backpla
ne Protocol, as enabling technology for using that mechanism. The idea unde
rlying the concept of the Internet Backplane is that NGI will enable us to
consider the global network as an extension of the processor backplane, if
only we have a low overhead mechanism for fine grained naming and access to
data, analogous to physical addresses and bus transfers. By this analogy t
he Internet Backplane is a common namespace for fine-grained management of
distributed resources. The IBP provides a flexible interface to enable this
functionality, allowing reliable and flexible control of remote storage bu
ffers through a general scheme for naming, staging, delivering and protecti
ng data.
We will test logistical QoS as an enabling technology for NGI computing usi
ng NetSolve. NetSolve is a software environment for networked computing des
igned to transform disparate computers and software components into a unifi
ed, easy-to-access computational service; it is being used by NSF's Partner
ships for Advanced Computational Infrastructure to build high-performance s
ystems for distributed computation on leading edge networks. We will invest
igate the implementation of logistical QoS within NetSolve and the integrat
ion of IBP with the Network Weather Service (used to monitor and forecast t
he performance of network and computational resources) to build a schedulin
g capability that maximizes the performance of NetSolve across next generat
ion networks. (C) 1999 Elsevier Science B.V. All rights reserved.