QoS as Middleware: Bandwidth Brokering System Design

We describe an approach to providing reservable bandwidth as a service for distributed computing environments such as computational Grids that aggregate resources to solve a single problem , remote instruments that operate on a schedule and that depend on communications and other Grid resources in order to function, etc. Our system provides the mechanisms for applications to make and use advance reservations for bandwidth. These mechanisms are built on top of the existing work in IP differentiated services. (The description in this paper is given in terms of IP differentiated service classes, but the intent is that it could also be applied to, e.g., a dynamic ATM circuit set up mechanism where QoS can be specified.) This document provides an overview of the bandwidth brokering system's model and design. It also describes the details of the components of the architecture, explains the flow of control among the components and describes typical use. The system performs reservation-based scheduling and supplies information to support rudimentary negotiation by the broker element. Our service works with current computational Grids by extending the Globus Toolkit. In particular, we use the Globus Resource Specification Language (RSL) to express reservation needs and the Globus GRAM interface to local resource managers for the service itself. Globus' DUROC will provide the initial broker service. The system uses the Akenti access control system to provide policy-based use of a class of service or a circuit based allocation. We describe our approach to providing quality of service in an IP-based (inter)network that supports some means of differentiating between classes of service. Our work will extend this infrastructure by providing a way to request and to confirm reservation of a service class end-to-end. Section 1.1 presents the system's target audience and its requirements. Section 1.2 describes how the required per-node resource reservation is performed using a slot scheduler, and Section 1.3 discusses access control for the reservation mechanism. Section 1.4 explains how a broker performs advanced, end-to-end bandwidth reservation using multiple slot schedulers, with Section 1.5 exploring the required access-control mechanism. Section 1.6 describes resource claiming. 1.1 Requirements of users We are addressing QoS tailored to serve a rather specialized audience. One type of user will be scientists who are performing experiments on singular instruments such as the LBL Advanced