Amaranth Summary Information

6/27/97

Objective:

The Amaranth project will provide multi-dimensional, adaptive, assured Quality of Service (QoS) for heterogeneous distributed computing systems. In particular, Amaranth will provide probabilistic guarantees of service encompassing real-time deadlines, dependability, cryptographic security, and application-specific performance. Amaranth will incorporate adaptive, distributed optimization techniques to maximize system performance even in degraded operating modes. User-specific tradeoffs of service commitment duration, commitment firmness, and service dimensions will provide appropriate service to all users even in a resource-limited system. A specific goal is to provide an order of magnitude improvement in system efficiency over static resource allocation, while assuring a comparable QoS.

Approach:

Amaranth will combine elements of theory, measurement, negotiation, resource management, and system-level prototyping. Four QoS dimensions (real-time deadlines, dependability, cryptographic security, and application-specific performance) will be related to system resources to enable significant ranges of resource/performance tradeoffs. Visualization and performance modeling tools will support viewing and changing QoS requests for the application, and tuning system QoS negotiation policies for efficiency.

System performance will be dynamically adjusted under varying workloads and resource levels. This will be achieved through the development of theoretical models of end-to-end QoS requirements and negotiation, as well as via the creation of performance monitoring capabilities and measurement tools. In particular, our work will have the following innovations in the underlying areas:

• Theory: Amaranth will formally express different dimensions of end-to-end QoS requirements within a unified theoretical framework. Then, a two-tiered modeling approach for decentralized resource usage optimization will be implemented, incorporating both analytic admission control and decentralized resource allocation.
• Measurement: Amaranth will define portable and adaptable QoS parameters for operating system and application software interfaces. Resource usage will be calculated and predicted using these system-independent parameters.
• Negotiation: Amaranth will incorporate per-node scheduling approaches based on analytic models and local measurements to assure a requested level of QoS. Inter-node dynamic optimization will be accomplished with adaptive negotiation algorithms for system-level QoS capabilities. Optimization will consider the semantic importance of different datastreams and will be achieved in a completely decentralized manner.
• Resources: Amaranth will incorporate tools for both single- and multi-dimensional visualization and tuning of QoS versus resource consumption.
• System: In order to provide a concrete framework for developing Amaranth capabilities, it will be implemented with a multi-session, tactically critical video communication application upon a real-time network of workstations. This testbed will be used to observe and tune QoS tradeoffs under varying workloads and resource availabilities. Quantification of the effectiveness of the decentralized approach in terms of efficiency and ability to meet QoS commitments will be accomplished by comparing it to a comparable statically scheduled system.

FY 1998 Plan:

• Create a multi-dimensional QoS theoretical framework for inter-dimensional tradeoffs. This framework will encompass the notion of time-sensitive, probabilistic QoS commitments to accommodate the reality of system degradation and overloads.
• Create resource models/simulations to provide a representative QoS workload for quantifying later project results .
• Develop an initial version of a single-dimension QoS visualization tool for instrumenting and understanding later results.
• Select an initial multi-agent based single-dimensional negotiation strategy for QoS optimization. Special emphasis will be placed on decentralized approaches that can survive equipment failures and overloads.

Amaranth is conducted by the Institute for Complex Engineered Systems (ICES) at Carnegie Mellon University. Amaranth is sponsored by DARPA ITO, Quorum program, under contract [number TBD]. The contents of this web site are the product of Amaranth researchers, and are not necessarily endorsed by DARPA. Web site by Jeff Hansen. Comments to: hansen+@ices.cmu.edu. For problems or questions regarding this web contact stamboli@andrew.cmu.edu. Last updated: May 13, 1998.