Multimedia servers present time and resource requirements that differ
significantly from those of conventional file servers.  
In particular, multimedia information, which includes audio and motion
video, requires a guaranteed transfer rate and large storage capacity.
The I/O subsystem is therefore a performance and cost bottleneck.

In this talk I will survey recent results in the study of performance 
issues in multimedia I/O subsystems: a main challenge is to effectively 
balance the load on the storage devices (e.g., disks) comprising the 
I/O subsystems, so as to maximize the throughput the system can achieve. 

I will present a general scheme which balances the load by combining 
efficient storage management with real-time disk scheduling. The scheme is
based on an initial static data placement phase, followed by a dynamic
phase, in which the load on the disks is balanced while "playing" multiple
streams of different video files simultaneously. 

The overall performance of the I/O subsystem strongly depends on the
quality of the initial placement of the data on the disks.
Several measures for this quality may be considered.
I will describe the resulting optimization problems, and how they 
can be solved efficiently, using algorithms for class-constrained packing.