1 / 1

RED-PD: RED with Preferential Dropping Ratul Mahajan Sally Floyd (ACIRI) David Wetherall

CSFQ. FRED. SFB. RED. Preferential Dropping Approaches. RED-PD: RED with Preferential Dropping Ratul Mahajan Sally Floyd (ACIRI) David Wetherall. RED-PD identifies and controls high bandwidth flows at the congested router to improve fairness.

tavi
Download Presentation

RED-PD: RED with Preferential Dropping Ratul Mahajan Sally Floyd (ACIRI) David Wetherall

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. CSFQ FRED SFB RED Preferential Dropping Approaches RED-PD: RED with Preferential Dropping Ratul Mahajan Sally Floyd (ACIRI) David Wetherall • RED-PD identifies and controls high bandwidth flows at the congested router to improve fairness. Challenge:to do the above while keeping routers simple. No flowstate Per-flow state High bandwidth flow state Misbehaving flow state • Spectrum of possibilities • No flow state is simple but implies no protection. • Per-flow state provides protection but is heavyweight in presence of lots of “Web mice”. • RED-PD combines simplicity and protection by keeping state only for flows which are high bandwidth. We call this “partial flow state”. Continuum of policies FQ CBQ Router Mechanisms FIFO SFQ DRR Scheduling Approaches RED-PD Why the partial flow state approach works? • RED-PD internals • Define target bandwidth T as throughput of a TCP flow with RTT R given drop rate p at the router. • Use drop history of length dependent on R and p to identify flows sending more than T. Divide the history into multiple lists to filter out losses in a window of data. • Preferentially drop from these flows with a dropping probability based on the flows’ sending rate. Make iterative changes to dropping probability based on p and feedback from identification engine to bring down the flow to T before inserting it into the output queue. Bandwidth distribution over different time windows at a router. • Bandwidth distribution is heavy tailed. A small fraction of flows account for most of the bandwidth. • In this case, controlling only 10% of the flows will give control over 80% of the bandwidth. Effect of RED-PD • Flows above T are preferentially dropped, and state kept for them. • Decreasing T increases fairness, but also state. Main features • Enhanced fairness using much less state. • Tunable degree of fairness. • Reasonable response time. Simulation Results More simulations done to evaluate RED-PD for response time, web traffic, TFRC, multiple congested links and effect of R Flows 1-9 are TCPs, 10-12 are high bandwidth CBRs. • Future directions • Identify unresponsive flows and actively punish them. • Investigate state requirements in more detail. http://www.cs.washington.edu/homes/ratul

More Related