Performability: An e-utility Imperative
Abstract
An e-business infrastructure is emerging to
support a new generation of server-hosted, network-delivered
applications and services: the e-utility. An e-utility is
implemented by scalable clusters of small servers which,
collectively, deliver Web content and services for multiple
businesses. Demands for such services, as experienced by an entire
utility, typically consist of concurrent and independent read-only
requests, resulting in a randomly varying workload whose peak rate
far exceeds the average request-rate. Hence, significant
overprovisioning of capacity is currently quite common. This
situation is undesirable for the utility provider and differential
quality of service (QoS) is emerging as a possible remedy. Over
time, the allocation of computing resources may change from one
business to another, depending on prior contractual agreements
(e.g.,workload prioritization, resource limits) and the availability
of other servers (as randomly affected by faults, local workloads,
etc.). Hence, dynamics of both an e-utility and its operational
environment can substantially affect QoS perceived by users.
To evaluate e-utility QoS, we contend that the use of model-based
performability evaluation techniques and tools is imperative. Our
discussion begins with some relevant background regarding both
e-utilities and the concept of performability. The above contention is
then supported with a more general argument
based on definitions of QoS provided by the telecom and Internet
communities. Specifically, we show that performability measures
specialize to QoS measures and, moreover, provide an effective means
of expressing the "collective effect" of lower level measures. A
performability approach to evaluating e-utility QoS is then
discussed, where the main steps concern QoS measure specification,
workload characterization, system model construction, measure
formulation (in terms of model behavior), and finally measure
evaluation.
Compressed PDF File: icssea.pdf.zip