1 / 14

EE689 Lecture 5

EE689 Lecture 5. Review of last lecture More on HPF RED. Network Congestion. Congestion results in buffer backlog Eventually, packets have to be dropped Packet drops indicate congestion to senders Senders eventually backoff to reduce congestion

nieve
Télécharger la présentation

EE689 Lecture 5

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. EE689 Lecture 5 • Review of last lecture • More on HPF • RED

  2. Network Congestion • Congestion results in buffer backlog • Eventually, packets have to be dropped • Packet drops indicate congestion to senders • Senders eventually backoff to reduce congestion • Usual mode of dropping packets at router - DropTail

  3. Problems with DropTail • No indication of congestion until late • May drop several packets at once - leads to global synchronization of flows • Can lead to oscillations of high-low link utilization • Aggressive flows can kill all other flows

  4. Early Congestion Indication • Notify congestion earlier • Individual flows reduce rates at different times -- reducing synchronization problems • Various techniques for early congestion indication/notification • Random Early Drop -- drop packets randomly when queues are starting to buildup

  5. Random Early Detection • Two thresholds, min, max of queue lengths • In between , mark/drop packets randomly • Inform flows early on, avoid global synchronization • Reduces average queue lengths, delays • Flows with higher rates have more packets dropped - fairer distribution of BW

  6. RED For each packet arrival, calculate the average queue size avg if min  avg < max calculate probability pa, mark arriving packet with probability pa else if max  avg, mark the arriving packet

  7. RED • Pb = max(avg - min)/(max -min) • pa = pb/(1 - count*pb) • avg = (1-w)*avg + w * q • count measures the number of packet arrivals since the last marked packet if queue length is between min and max

  8. RED Drop p Queue length

  9. RED properties • Maintains queue lengths low • Allows high utilization of links without oscillations/global synchronization • Higher rate flows get dropped more often -leads to fairer sharing than DropTail • Aggressive flows can still shut down ‘nice’ applications • Small RTT bias of TCP still present

  10. RED properties • FIFO scheduling - doesn’t provide flow isolation - one aggressive flow can hog all the buffers • Favors ‘robust’ applications (ftp) over ‘fragile’ applications (telnet) • FRED - Per Flow RED employs RED and per-flow queuing to solve some of these problems

  11. Buffer management • FIFO scheduling cannot isolate flows • Fair scheduling shown to contain flows • Can fail without appropriate buffer management -- aggressive flows can eat up all the buffers and not let other flows have any backlog • Buffer management important for flow isolation

  12. Buffer management • Fair scheduling doesn’t guarantee flow isolation -- aggressive flows can hog all the buffers • Per-flow queuing is required, contain flows from using too many buffers • Fair sharing of buffers leads to fair sharing of bandwidth

  13. Fair sharing of BW • Drop packets of longest queue • Leads to more losses for aggressive flows • Contains aggressive flows • Leads to fair sharing of BW even without aggressive flows • Possible to achieve BW targets by strict buffer control, assigns buffers proportional to BW targets

  14. Summary • RED provides fairer sharing than DropTail • RED simple to implement • RED cannot contain aggressive flows • Per-flow queuing also not enough • Buffer management required • Effective buffer management can lead to fair sharing of BW even with FIFO scheduling

More Related