The flow scheduling problem is a recurrent issue in computer networking, stemming from the need for optimal resource sharing among users, devices, threads or services. Prominent examples include the packet routing, Medium Access Control and thread-to-CPU assignment problems. In a context-agnostic manner, flow scheduling is the definition of a service pattern for assigning "jobs" to "processors", while adhering to job-specific "conditions".

Treating jobs per unit cannot lead to persistent, optimal service patterns, unless NP=P.

However, aggregating jobs into larger-scale "job-flows" with persistent statistics yields tractability. The optimal service pattern becomes periodical, while an unbounded number of job conditions can be incorporated to the process in a modular fashion. This feature is dubbed as "multi-optimal" flow scheduling. Real-world applications, such as flow scheduling on an optical router, may pose the additional restriction of minimal computational complexity of the scheduling process. Thus, the talk will outline a novel method that makes the periodic scheduling process parametrical to its complexity.

   Finally, the talk will conclude with a short overview of studies on adaptive flow scheduling, job aggregation techniques based on web measurements and related security issues. Prospects of applications to the NetVolution project will be presented as discussion topics.