Resource management method and apparatus for maximizing multimedia performance of open systems

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A meaningful definition for QOS in MM communications must encompass the entire solution architecture rather than be limited to disk I/O or server or network.
Taking the lead from business computing and communication arena, it is reasonable for a MM customer to desire: (a) response time for the continuous media requests to be bounded by a reasonable time; and (b) service for all admitted clients to be continuous for a period with the requested quality in terms of Mbit rates, delay and jitter.
The QOS for MM is determined by the response time and data delivery rate at the client.
This is influenced by the processing scenario within the system incorporating the information source (e.
g.
a disk subsystem), network connections from other servers or clients, and the software.
GQOS within a system is possible under the following conditions: 1.
the system is configured properly meeting the customer's stated needs; 2.
admission control is based on current values of (monitored or model predicted) performance parameters for key components and the impact of new request admission on the QOS of new and existing requests; and 3.
protection against viruses and misbehaving applications whose resource demands hurt QOS for all.
An obvious solution is to design servers that provide GQOS from their part and couple them with the ATM network that has QOS built by design so that the entire system provides GQOS for MM customers.
However, proprietary real time computers are more expensive than off-the-shelf open systems.
Further, users might prefer emerging open system computers that include symmetric multiprocessing, highly parallel architectures, operating systems, databases, high availability and other technologies.
The present invention provides an approach that enables an open system server to support GQOS.
Admission control has received considerable attention in the art.
One proposed technique provides an observation based method that assumes that the average amount of time spent in retrieving a media block from disk does not change significantly even after the new client is admitted by the server