Networking Acronym Smorgasbord: , DVMRP, CBT, WFQ

Transcription

1 Networking Acronym Smorgasbord: , DVMRP, CBT, WFQ EE122 Fall 2011 Scott Shenker Materials with thanks to Jennifer Rexford, Ion Stoica, Vern Paxson and other colleagues at Princeton and UC Berkeley 1

2 Announcements Congratulations: You all got 100% on HW4 Worksheet will provide practice This is last week of sections See posting about additional office hours next week Next week will have office hours during class times Will work through problems on work sheet Be there or be square. Wednesday s Review: will figure something out. 2

3 Today s Lecture: Dim Sum of Design Wireless review Multicast Packet Scheduling Peer-to-peer 3

4 Wireless Review 4

5 History MACA proposal: basis for RTS/CTS in lecture Contention is at receiver, but CS detects sender! Replace carrier sense with RTS/CTS MACAW paper: extended and altered approach Implications of data ACKing Introducing DS in exchange: RTS-CTS-DS-Data-ACK o Shut up when hear DS or CTS Other clever but unused extensions for fairness, etc : uses carrier sense and RTS/CTS RTS/CTS often turned off, just use carrier sense When RTS/CTS turned on, shut up when hear either RTS/CTS augments carrier sense 5

6 What Will Be on the Final? General awareness of wireless (lecture) Reasoning about a given protocol If we used the following algorithm, what would happen? You are not expected to know which algorithm to use; we will tell you explicitly. 6

7 Multicast 7

8 Motivating Example: Internet Radio Internet concert More than 100,000 simultaneous online listeners Could we do this with parallel unicast streams? Bandwidth usage If each stream was 1Mbps, concert requires > 100Gbps Coordination Hard to keep track of each listener as they come and go Multicast addresses both problems. 8

9 Unicast approach does not scale Broadcast Center Backbone ISP 9

10 Instead build data replication trees Copy data at routers At most one copy of a data packet per link Broadcast Center Backbone ISP LANs implement link layer multicast by broadcasting Routers keep track of groups in real-time Routers compute trees and forward packets along them 10

11 Multicast Service Model R 0 S [G, data] Net R 1... R n Receivers join multicast group identified by a multicast address G Sender(s) send data to address G Network routes data to each of the receivers Note: multicast is both a delivery and a rendezvous mechanism Senders don t know list of receivers For many purposes, the latter is more important than the former 11

12 Multicast and Layering Multicast can be implemented at different layers link layer o e.g. Ethernet multicast network layer o e.g. IP multicast application layer o e.g. End system multicast Each layer has advantages and disadvantages Link: easy to implement, limited scope IP: global scope, efficient, but hard to deploy Application: less efficient, easier to deploy [not covered] 12

13 Multicast Implementation Issues How is join implemented? How is send implemented? How much state is kept and who keeps it? 13

14 Link Layer Multicast Join group at multicast address G NIC normally only listens for packets sent to unicast address A and broadcast address B After being instructed to join group G, NIC also listens for packets sent to multicast address G Send to group G Packet is flooded on all LAN segments, like broadcast Scalability: State: Only host NICs keep state about who has joined Bandwidth: Requires broadcast on all LAN segments Limitation: just over single LAN 14

15 Network Layer (IP) Multicast Performs inter-network multicast routing Relies on link layer multicast for intra-network routing Portion of IP address space reserved for multicast 2 28 addresses for entire Internet Open group membership Anyone can join (sends IGMP message) o Internet Group Management Protocol Privacy preserved at application layer (encryption) Anyone can send to group Even nonmembers 15

16 How Would YOU Design this? 5 Minutes. 16

17 IP Multicast Routing Intra-domain (know the basics here) Source Specific Tree: Distance Vector Multicast Routing Protocol (DVRMP) Shared Tree: Core Based Tree (CBT) Inter-domain [not covered] Protocol Independent Multicast Single Source Multicast 17

18 Distance Vector Multicast Routing Protocol Elegant extension to DV routing Using reverse paths! Use shortest path DV routes to determine if link is on the source-rooted spanning tree See whiteboard.. Three steps in developing DVRMP Reverse Path Flooding Reverse Path Broadcasting Truncated Reverse Path Broadcasting (pruning) 18

19 Reverse Path Flooding (RPF) If incoming link is shortest path to source Send on all links except incoming Otherwise, drop s:3 s:2 s:3 s:1 s:2 Issues: (fixed with RPB) s r Some links (LANs) may receive multiple copies Every link receives each multicast packet 19

20 Other Problems Flooding can cause a given packet to be sent multiple times over the same link S x y a z duplicate packet b Solution: Reverse Path Broadcasting 20

21 Reverse Path Broadcasting (RPB) Choose single parent for each link along reverse shortest path to source Only parent forwards to child link Identifying parent links Distance Lower address as tiebreaker Parent of z on reverse path forward only to child link child link of x for S a b x S 5 6 y z 21

22 Even after fixing this, not done This is still a broadcast algorithm the traffic goes everywhere Need to Prune the tree when there are subtrees with no group members Networks know they have members based on IGMP messages Add the notion of leaf nodes in tree They start the pruning process 22

23 Pruning Details Prune (Source,Group) at leaf if no members Send Non-Membership Report (NMR) up tree If all children of router R send NMR, prune (S,G) Propagate prune for (S,G) to parent R On timeout: Prune dropped Flow is reinstated Down stream routers re-prune Note: a soft-state approach 23

24 Distance Vector Multicast Scaling State requirements: O(Sources Groups) active state How to get better scaling? Hierarchical Multicast Core-based Trees 24

25 Core-Based Trees (CBT) Pick rendevouz point for the group (called core) Build tree from all members to that core Shared tree More scalable: Reduces routing table state from O(S x G) to O(G) 25

26 Use Shared Tree for Delivery Group members: M1, M2, M3 M1 sends data root M1 M2 M3 control (join) messages data 26

27 Barriers to Multicast Hard to change IP Multicast means changes to IP Details of multicast were very hard to get right Not always consistent with ISP economic model Charging done at edge, but single packet from edge can explode into millions of packets within network 27

28 Packet Scheduling 28

29 Scheduling Decide when and what packet to send on output link Classifier partitions incoming traffic into flows In some designs, each flow has their own FIFO queue flow 1 1 Classifier flow 2 Scheduler 2 flow n Buffer management 29

30 Packet Scheduling: FIFO What if scheduler uses one first-in first-out queue? Simple to implement But everyone gets the same service Example: two kinds of traffic Video conferencing needs low bandwidth and low delay o E.g., 1 Mbps and 100 msec delay not sensitive to delay, but need bandwidth Cannot admit much traffic Since it will interfere with the video conference traffic 30

31 Packet Scheduling: Strict Priority Strict priority Multiple levels of priority Always transmit high-priority traffic, when present.. and force the lower priority traffic to wait Isolation for the high-priority traffic Almost like it has a dedicated link Except for the (small) delay for packet transmission o High-priority packet arrives during transmission of low-priority o Router completes sending the low-priority traffic first 31

32 Scheduling: Weighted Fairness Limitations of strict priority Lower priority queues may starve for long periods even if the high-priority traffic can afford to wait Traffic still competes inside each priority queue Weighted fair scheduling Assign each queue a fraction of the link bandwidth Rotate across the queues on a small time scale Send extra traffic from one queue if others are idle 50% red, 25% blue, 25% green 32

33 Max-Min Fairness Given a set of bandwidth demands r i and a total bandwidth C, the max-min bandwidth allocations are: a i = min(f, r i ) where f is the unique value such that Sum(a i ) = C Property: If you don t get full demand, no one gets more than you 33

34 Computing Max-Min Fairness Denote C link capacity N number of flows r i arrival rate Max-min fair rate computation: 1. compute C/N (= the remaining fair share) 2. if there are flows i such that r i C/N then update C and N C = C r i s.t r i ; N = N k (for k such flows) i C / N and go to 1 3. if not, f = C/N; terminate 34

35 Example C = 10; r 1 = 8, r 2 = 6, r 3 = 2; N = 3 C/3 = 3.33 Can service all of r 3 Remove r 3 from the accounting: C = C r 3 = 8; N = 2 C/2 = 4 Can t service all of r 1 or r 2 So hold them to the remaining fair share: f = f = 4: min(8, 4) = 4 min(6, 4) = 4 min(2, 4) = 2 35

36 Fair Queuing (FQ) Conceptually, computes when each bit in the queue should be transmitted to attain max-min fairness (a fluid flow system approach) Then serve packets in the order of the transmission time of their last bits Allocates bandwidth in a max-min fairly 36

37 Example Flow 1 (arrival traffic) time Flow 2 (arrival traffic) time Service in fluid flow system time Packet system time 37

38 Fair Queuing (FQ) Provides isolation: Misbehaving flow can t impair others Could change congestion control paradigm o But not used. Doesn t solve congestion by itself: Still need to deal with individual queues filling up Generalized to Weighted Fair Queuing (WFQ) Can give preferences to classes of flows Used for quality of service (QoS) o Allocations to aggregates 38