Three new approaches for adjustment and improvement of the RR scheduler in a dynamic resource environment

  • Authors:
  • John Tsiligaridis;Raj Acharya

  • Affiliations:
  • Computer Science and Engineering, 236 Bell Hall, Buffalo, NY 14260 SUNY at Buffalo, USA;Computer Science and Engineering, 220 Pond Laboratory, Penn. State University, University Park, PA 16802-6106, USA

  • Venue:
  • Computer Communications
  • Year:
  • 2005

Quantified Score

Hi-index 0.24

Visualization

Abstract

Many methods of packets' service have been developed, such as the Deficit Round Robin (DRR) and the Surplus Round Robin (SRR), that have to do with a borrowing of credits technique for use in the next rounds. In the present work, we propose a new variable flexible credit scheme, for adjustment of the weights, instead of a fixed credit scheme with a lower maximum weight. The service of various short duration flows with large size packets (impossible to be serviced in a round), is achieved by using gradual or direct weight increase mechanisms. A feedback mechanism is used for adjusting the weight size taking into account the parameter of the current packet size. This constitutes an alternative way of defining the most suitable size of the quantum versus the traditional one of the weight's computation (like SLA). The advantage of our best weight adjusting method lies on both the restriction of the otherwise required, repeated, unexpected and uncertain changes of the next round's credits (as it happens with the uninformed for the large packets arrival Deficit Round Robin) and the guarantee of the next packet's service with no extra overhead ('empty round'). With this ability, the DRR has again O(1) complexity, without making any assumption for the maximum size of the packets. A family of algorithms, with O(1) complexity, that work on the fly, using different and systematic weight request methods are: the Direct Increasing Weight (DIW), the Compound Round Robin (CRR) and the Proportional Increasing Weight (PIW). In the case of large packets sequence's arrival, our algorithms outclasses the low static quantum DRR. Upgrading the DRR using the PIW, is a different approach that outperforms the DRR (fixed weight). Their relative fairness along with the delay bounds are derived. Simulation experiments support the algorithms' significance and useful results are also provided in comparison with other scheduling algorithms. This kind of algorithms can be applied for next generation internet routers, achieving short time service for delay sensitive flows.