EE 122: Router Support for Congestion Control: RED and Fair Queueing. Ion Stoica Oct. 30 Nov. 4, 2002

Transcription

1 EE 122: Router Support for Congestion Control: RED and Fair Queueing Ion Stoica Oct. 30 Nov. 4, 2002

2 Router Support For Congestion Management Traditional Internet - Congestion control mechanisms at end-systems, mainly implemented in TCP - Routers play little role Router mechanisms affecting congestion management - Scheduling - Buffer management Traditional routers - FIFO - Tail drop istoica@cs.berkeley.edu 2

3 Router Packet Processing Scheduling: - Decide when to send a packet - Decide which packet to transmit Buffer management: - Decide when to drop a packet - Decide which packet to drop istoica@cs.berkeley.edu 3

4 FIFO: First-In First-Out Maintain a queue to store all packets When to send? - When there is at least one packet in the queue Which packet to send? - The packet at the head of the queue Next to transmit Arriving packet Queued packets istoica@cs.berkeley.edu 4

5 Tail-drop Buffer Management When to drop? - When buffer full Which packet to drop - The arriving packet Next to transmit Arriving packet Drop istoica@cs.berkeley.edu 5

6 Drawbacks of FIFO with Tail-drop Synchronizing effect for multiple TCP flows Buffer lock out by misbehaving flows Burst or multiple consecutive packet drops 6

7 Synchronizing effect for multiple TCP flows When buffer overflows many/all flows experience losses Flows which experience losses will scale down their transmission (by reducing the window size) This will cause the link to be underutilized the process repeats Net result: inefficient link utilization 7

8 Buffer Lock Out by Misbehaving Flows Throughput(Mbps) Aggressive flows can squeeze out TCP flows Example: one 10 Mbps link shared by UDP - 1 UDP flow that sends at 10 Mbps, and does not implement any congestion control - 31 TCP flows FIFO Flow Number UDP (#1) - 10 Mbps TCP (#2) TCP (#32) Bottleneck link (10 Mbps) istoica@cs.berkeley.edu 8 UDP (#1) TCP (#2) TCP (#32)

9 Burst of Packet Losses A TCP flow sends a burst of packets that arrive at a router when buffer is full All packets in the burst are lost This may cause RTO to trigger: cwnd 1 For example, with TCP Reno if there are three packet losses in a window, this will almost sure trigger an RTO istoica@cs.berkeley.edu 9

10 FIFO Router with Two TCP Flows 10

11 Random Early Detection (RED) Routers (Floyd & Fall 93] Goals - Avoid synchronizing multiple TCP flows - Avoid burst of packet drops istoica@cs.berkeley.edu 11

12 FIFO scheduling Buffer management: RED - Probabilistically discard packets - Probability is computed as a function of average queue length (why average?) Discard Probability 1 0 min_th max_th queue_len Average Queue Length istoica@cs.berkeley.edu 12

13 RED (cont d) min_th minimum threshold max_th maximum threshold avg_len average queue length - avg_len = (1-w)*avg_len + w*sample_len Discard Probability 1 0 min_th max_th queue_len Average Queue Length istoica@cs.berkeley.edu 13

14 RED (cont d) If (avg_len < min_th) enqueue packet If (avg_len > max_th) drop packet If (avg_len >= min_th and avg_len < max_th) enqueue packet with probability P Discard Probability (P) 1 0 min_th max_th queue_len Average Queue Length istoica@cs.berkeley.edu 14

15 RED (cont d) P = max_p*(avg_len min_th)/(max_th min_th) Improvements to spread the drops (see textbook) Discard Probability max_p P 1 0 min_th max_th queue_len Average Queue Length avg_len istoica@cs.berkeley.edu 15

16 RED Advantages Absorb burst better Avoids synchronization Signal end systems earlier 16

17 RED Router with Two TCP Sessions 17

18 Problems with RED Throughput(Mbps) No protection: if a flow misbehaves it will hurt the other flows Example: 1 UDP (10 Mbps) and 31 TCP s sharing a 10 Mbps link UDP RED Flow Number UDP (#1) - 10 Mbps TCP (#2) TCP (#32) Bottleneck link (10 Mbps) istoica@cs.berkeley.edu 18 UDP (#1) TCP (#2) TCP (#32)

19 Solution? Round-robin among different flows [Nagle 87] - one queue per flow istoica@cs.berkeley.edu 19

20 Round-Robin Discussion Advantages: protection among flows - Misbehaving flows will not affect the performance of wellbehaving flows Misbehaving flow a flow that does not implement any congestion control - FIFO does not have such a property Disadvantages: - More complex than FIFO: per flow queue/state - Biased toward large packets a flow receives service proportional to the number of packets (When is this bad?) istoica@cs.berkeley.edu 20

21 Solution? Bit-by-bit round robin Can you do this in practice? No, packets cannot be preempted (why?) we can only approximate it 21

22 Fair Queueing (FQ) [DKS 89] Define a fluid flow system: a system in which flows are served bit-by-bit Then serve packets in the increasing order of their deadlines Advantages - Each flow will receive exactly its fair rate Note: - FQ achieves max-min fairness istoica@cs.berkeley.edu 22

23 Denote - C link capacity - N number of flows - r i arrival rate Max-Min Fairness Max-min fair rate computation: 1. compute C/N 2. if there are flows i such that r i <= C/N, update C and N C = C 3. if no, f = C/N; terminate 4. go to 1 i s. t r C / N A flow can receive at most the fair rate, i.e., min(f, r i ) i r i istoica@cs.berkeley.edu 23

24 Example C = 10; r 1 = 8, r 2 = 6, r 3 = 2; N = 3 C/3 = 3.33 C = C r3 = 8; N = 2 C/2 = 4; f = 4 f = 4 min(8, 4) = 4 min(6, 4) = 4 min(2, 4) = 2 istoica@cs.berkeley.edu 24

25 Implementing Fair Queueing Idea: serve packets in the order in which they would have finished transmission in the fluid flow system 25

26 Example Flow 1 (arrival traffic) time Flow 2 (arrival traffic) time Service in fluid flow system time Packet system time istoica@cs.berkeley.edu 26

27 System Virtual Time: V(t) Measure service, instead of time V(t) slope rate at which every active flow receives service - C link capacity - N(t) number of active flows in fluid flow system at time t V(t) V ( t) t = C N ( t) time Service in fluid flow system time istoica@cs.berkeley.edu 27

28 Fair Queueing Implementation Define k F i - - finishing time of packet k of flow i (in system virtual time reference system) k a i k L i - - arrival time of packet k of flow i - - length of packet k of flow i The finishing time of packet k+1 of flow i is F k + 1 i = max( V ( a k i + 1 ), F k i ) + L k + 1 i istoica@cs.berkeley.edu 28

29 FQ Advantages FQ protect well-behaved flows from ill-behaved flows Example: 1 UDP (10 Mbps) and 31 TCP s sharing a 10 Mbps link Throughput(M bps) RED Throughput(M bps) FQ Flow Number Flow Number istoica@cs.berkeley.edu 29

30 Big Picture FQ does not eliminate congestion it just manages the congestion You need both end-host congestion control and router support for congestion control - end-host congestion control to adapt - router congestion control to protect/isolate Don t forget buffer management: you still need to drop in case of congestion. Which packet s would you drop in FQ? - one possibility: packet from the longest queue istoica@cs.berkeley.edu 30