Network Virtualization and Data Center Networks Introduction

Transcription

1 Network Virtualization and Data Center Networks Introduction Qin Yin Fall Semester 2013 With thanks to Jennifer Rexford 1

2 WHAT IS NETWORKING? 2

3 A Plethora of Protocol Acronyms? BGP ARP HTTP DNS PPP OSPF DHCP TCP UDP SMTP FTP SSH MAC IP RIP NAT CIDR VLAN VTP NNTP POP IMAP RED ECN SACK SNMP TFTP TLS WAP SIP IPX STUN RTP RTSP RTCP PIM IGMP ICMP MPLS LDP HIP LISP LLDP BFD 3

4 A Heap of Header Formats? 4

5 TCP/IP Header Formats in Lego 5

6 A Big Bunch of Boxes? Router Deep Packet Inspection NAT Gateway WAN accelerator Label Switched Router Firewall Intrusion Detection System DNS server Hub Load balancer Base station Bridge DHCP server Scrubber Packet sniffer Route Reflector Switch Packet shaper Proxy Repeater 6

7 A Ton of Tools? arpwatch syslog traceroute nslookup nmap rancid net-snmp dig snort whois ntop ping tcpdump ipconfig iperf trat wget bro NDT dummynet wireshark mrtg 7

8 AN APPLICATION DOMAIN? 8

9 Application Domain for Theory? Algorithms and data structures Control theory Queuing theory Optimization theory Game theory and mechanism design Formal methods Information theory Cryptography Programming languages Graph theory 9

10 Application Domain for Systems? Distributed systems Operating systems Computer architecture Software engineering 10

11 So, Why is Networking Cool? (I) Tangible, relates to reality Can measure/build things (we do love our artifacts ) Can truly effect far-reaching change in the real world Inherently interdisciplinary Well-motivated problems + rigorous solution techniques Interplay with policy, economics, and social science Widely-read papers Many of the most cited papers in CS are in networking Congestion control, distributed hash tables, resource reservation, self-similar traffic, multimedia protocols, Three of top-ten CS authors (Shenker, Jacobson, Floyd) So, somebody is interested in reading this stuff 11

12 So, Why is Networking Cool? (II) Young, relatively immature field Great if you like to make order out of chaos Tremendous intellectual progress is still needed You can help decide what networking really is Defining the problem is a big part of the challenge Recognizing a need, formulating a well-defined problem is at least as important as solving the problem Lots of platforms for building your ideas Testbeds: Emulab, PlanetLab, GENI, Grid systems: Globus Cloud computing platforms: Amazon EC2, Microsoft Azure, Programmability: OpenFlow/NOX, NetFPGA, Click Measurements: RouteViews, traceroute, Internet2, 12

13 ABOUT THIS CLASS 13

14 Goals To gain in-depth understanding of two hot topics in the field of networking research Network virtualization Data center networks To get some practice in the art of reading papers To gain some practical experience through programming assignments It is a big field, and we will focus on just a few topics 14

15 Prerequisite General knowledge of Networking System programming Relevant courses Operating Systems and Networks 15

16 Grade Lectures (English) 10% In-class participation Programming assignments 60% Exam (Oral, 15min, English) 30% 16

17 Lecture Outline (tentative) Introduction, material review Overlay networks (RON, Overcast, SON, i3) Testbeds: PlanetLab, VINI, Emulab, GENI Grid computing: Globus, VNET & Cloud computing: Amazon, Azure VNE research directions, network mapping Data center network evolution (book) Intra-data center communication: VLAN, vswitch, VRF (book) Inter-data center communication: VPN, MPLS, tunneling (book) Data center traffic measurements and characteristics Data center communication: optimization SDN: Openflow, FlowVisor Novel data center network architecture: Google, Facebook, VL2 17

18 Programming assignments Four assignments PA #1: Resource management overlay PA #2: Investigation of different virtualized platforms PA #3: Exploration of buffer bloat PA #4: Service defined networking Computing facilities PlanetLab, Amazon AWS, Microsoft Anzure Language: Python We DON T assume you know Python! 18

19 Course Material Course webpage Text book Data Center Virtualization Fundamentals: Understanding Techniques and Designs for Highly Efficient Data Centers with Cisco Nexus, UCS, MDS, and Beyond (some chapters) Many research articles referenced during the lectures 19

20 Contact With subject prefix: [virnet] Office Tuesday 1-2pm 20

21 HOW TO READ A PAPER S. KESHAV 21

22 You Spend a Lot of Time Reading Reading papers for grad classes (like this one!) Reviewing papers for conferences/journals Giving colleagues feedback on their papers Keeping up with work related to your research Staying broadly educated about the field Transitioning into a new research area Learning how to write better papers So, it is worthwhile to learn to read effectively. 22

23 Keshav s Three-Pass Approach: Step 1 A ten-minute scan to get the general idea Title, abstract, and introduction Section and subsection titles Conclusion Bibliography What to learn: the five C s Category: What type of paper is it? Context: What body of work does it relate to? Correctness: Do the assumptions seem valid? Contributions: What are the main research contributions? Clarity: Is the paper well-written? Decide whether to read further 23

24 Keshav s Three-Pass Approach: Step 2 A more careful, one-hour reading Read with greater care, but ignore details like proofs Figures, diagrams, and illustrations Mark relevant references for later reading Grasp the content of the paper Be able to summarize the main thrust to others Identify whether you can (or should) fully understand Decide whether to Abandon reading the paper in any greater depth Read background material before proceeding further Persevere and continue on to the third pass 24

25 Keshav s Three-Pass Approach: Step 3 Several-hour virtual re-implementation of the work Making the same assumptions, recreate the work Identify the paper s innovations and its failings Identify and challenge every assumption Think how you would present the ideas yourself Jot down ideas for future work When should you read this carefully? Reviewing for a conference or journal Giving colleagues feedback on a paper Understanding a paper closely related to your research Deeply understanding a classic paper in the field 25

26 Other tips for reading papers Read at the right level for what you need Work smarter, not harder Read at the right time of day When you are fresh, not sleepy Read in the right place Where you are not distracted, and have enough time Read actively With a purpose (what is your goal?) With a pen or computer to take notes Read critically Think, question, challenge, critique, 26

27 A BRIEF REVIEW 27

28 One Take on Define Networking Definition and placement of function What to do, and where to do it The division of labor Between the host, network, and management systems Across multiple concurrent protocols and mechanisms 28

29 Host-Network Division of Labor Packet switching Divide messages into a sequence of packets Headers with source and destination address Best-effort delivery Packets may be lost / corrupted /out of order Fixed end points with IP addresses host network host 29

30 Intermediate Transport Layer But, applications want efficient, accurate transfer of data in order, in a timely fashion Let the end hosts handle all of that (An example of the end-to-end argument ) Transport layer can optionally Retransmit lost packets Put packets back in order Detect and handle corrupted packets Avoid overloading the receiver 30

31 Protocol layers: Hosts vs. Routers host HTTP HTTP message host HTTP TCP TCP segment TCP router router IP IP packet IP IP packet IP IP packet IP Ethernet interface Ethernet interface SONET interface SONET interface Ethernet interface Ethernet interface 31

32 Layer Encapsulation Connection ID Source/Destination Link Address 32

33 The Narrow waist of IP FTP HTTP NV TFTP Applications TCP IP UDP Waist UDP TCP NET 1 NET 2 NET n Data Link Physical The Hourglass Model The waist facilitates interoperability 33

34 THE HOST 34

35 The Role of the End Host Network discovery and bootstrapping How does the host join the network? How does the host get an address? Interface to networked applications What interface to higher-level applications? How does the host realize that abstraction? Distributed resource sharing What roles does the host play in network resource allocation decisions? 35

36 Network discovery and bootstrapping Three kinds of identifiers Host Name IP Address MAC Address Who am I? Hard-wired: MAC address Static configuration: IP interface configuration Dynamically learned: IP address configured by DHCP Who are you? Hard-wired: IP address in a URL, or in the code Dynamically looked up: ARP or DNS 36

37 Network discovery and bootstrapping Dynamic Host Configuration Protocol (DHCP) Given a MAC address, assign a unique IP address and tell host other stuff about the Local Area Network To automate the bootstrapping process Address Resolution Protocol (ARP) Given an IP address, provide the MAC address To enable communication within the Local Area Network Domain Name System (DNS) Given a host name, provide the IP address Given an IP address, provide the host name 37

38 Interface to Applications - Sockets Applications communicate using sockets Message socket: unreliable message delivery Stream socket: reliable stream of bytes (like a file) User process socket Operating System User process socket Operating System 38

39 Two Main Transport Protocols User Datagram Protocol (UDP) Just provides demultiplexing and error detection Header fields: port numbers, checksum, and length Low overhead, good for query/response and multimedia Transmission Control Protocol (TCP) Adds support for a stream of bytes abstraction Retransmitting lost or corrupted data Putting out-of-order data back in order Preventing overflow of the receiver buffer Adapting the sending rate to alleviate congestion Higher overhead, good for most stateful applications 39

40 Resource Allocation Challenges Best-effort network easily becomes overloaded No mechanism to block excess calls Instead excess packets are simply dropped Examples Shared Ethernet medium: frame collisions Ethernet switches and IP routers: full packet buffers Quickly leads to congestion collapse Goodput Load congestion collapse Increase in load that results in a decrease in useful work done. 40

41 End Hosts Adjusting to Congestion End hosts adapt their sending rates In response to network conditions Shared Ethernet Carrier sense: wait for link to be idle Collision detection: listen while transmitting Exponential back-off: wait before retransmitting IP network Additive increase, multiplicative decrease Slow start, fast retransmit, etc. 41

42 DATA AND CONTROL PLANE 42

43 Split into Data vs. Control Plane Data plane: packets Handle individual packets as they arrive Forward, drop, or buffer Mark, shape, schedule, Control plane: events Track changes in network topology Compute paths through the network Reserve resources along a path Motivated by need for high-speed packet forwarding 43

44 Adding Management Plane (policies) Making the network run well Traffic reaches the right destination Traffic flows over short, uncongested paths Unwanted traffic is discarded Failure recovery happens quickly Routers don t run out of resources A control loop with the network Measure (sense): topology, traffic, performance, Control (actuate): configure control and data planes 44

45 Data, Control, and Management Planes Data Control Management Time-scale Packet (nsec) Event (10 msec to sec) Tasks Location Forwarding, buffering, filtering, scheduling Line-card hardware Routing, signaling Router software Human (min to hours) Analysis, configuration Humans or scripts 45

46 Data and Control Planes data plane Processor control plane Line card Line card Line card Switching Fabric Line card Line card Line card 46

47 Routing vs. Forwarding Routing: control plane Computing paths the packets will follow Routers talking amongst themselves Individual router creating a forwarding table Forwarding: data plane Directing a data packet to an outgoing link Individual router using a forwarding table 47

48 Routing Protocols I What does the protocol compute? What algorithm does the protocol run? Spanning-tree construction (Ethernet LAN) Link-state routing (OSPF and IS-IS) Distance vector routing (RIP and EIGRP) Path-vector routing (BGP) Source routing (IP source routing but always disabled) End-to-end signaling (MPLS with RSVP) 48

49 Routing Protocols II How do routers learn end-host locations? Learning/flooding Used in Ethernet LANs Injecting into the routing protocol Used in OSPF & IS-IS, especially in enterprise networks Dissemination using a different protocol Internal BGP (ibgp) used in backbone networks Directory server Used in some data centers 49

50 Routing Protocols Conclusion Routing is challenging Distributed computation Challenges with scalability and dynamics Many different solutions for different environments Ethernet LAN: spanning tree, MAC learning, flooding Enterprise: link-state routing, injecting subnet addresses Backbone: link-state routing inside, path-vector routing with neighboring domains, and ibgp dissemination Data centers: many different solutions, still in flux E.g., link-state routing or multiple spanning trees E.g., directory service or injection of subnets into routing protocol An active research area 50

51 Data Plane Streaming algorithms that act on packets Matching on some bits, taking a simple action at behest of control and management plane Wide range of functionality Forwarding Access control Mapping header fields Traffic monitoring Resource allocation: buffering, scheduling, shaping, marking Deep packet inspection 51

52 Packet Forwarding Control plane computes a forwarding table Maps destination address(es) to an output link Handling an incoming packet Match: destination address Switch: Match on Destination MAC IP routers: match on IP prefix Action: direct the packet to the chosen output link Switching fabric Directs packet from input link to output link 52

53 Packet Filtering: Access Control Five tuple for access control lists (ACLs) Source and destination IP addresses TCP/UDP source and destination ports Protocol (e.g., UDP vs. TCP) Can be more sophisticated E.g., block all TCP SYN packets from outside hosts Should arriving packet be allowed in? Departing packet let out? 53

54 Mapping Header Fields Remap IP addresses and TCP/UDP port numbers Addresses: between end-host and NAT addresses Port numbers: to ensure each connection is unique Create table entries as packets arrive Src , Sport 1024, Dest , Dport 80 Map to Src , Sport 1024, Dest , Dport 80 Src , Sport 1024, Dest , Dport 80 Map to Src , Sport 1025, Dest , Dport 80 Challenges When to remove the entries Running services behind a NAT What if both ends of a connection are behind NATs 54

55 Passive Traffic Monitoring Counting the traffic Match based on fields in the packet header and update a counter of #bytes and #packets Examples Link IP prefixes TCP/UDP ports Individual flows Challenges Dest Prefix #Packets #Bytes / / / / Identify traffic aggregates in advance vs. reactively Summarizing other information (e.g., time, TCP flags) Not knowing if you see all packets in a connection 55

56 Buffering Drop-tail FIFO queue Packets served in the order they arrive and dropped if queue is full Random Early Detection (RED) When the buffer is nearly full drop or mark some packets to signal congestion Multiple classes of traffic Separate FIFO queue for each flow or traffic class with a link scheduler to arbitrate between them 56

57 Link Scheduling Strict priority Assign an explicit rank to the queues and serve the highest-priority backlogged queue Weighted fair scheduling Interleave packets from different queues in proportion to weights 50% red, 25% blue, 25% green 57

58 Traffic Shaping Force traffic to conform with a profile To avoid congesting downstream resources To enforce a contract with the customer Leaky-bucket shaping Can send at rate r and intermittently burst Parameters: token rate r and bucket depth d Tokens arrive (rate r) Max # of tokens (d tokens) 58 packets tokens A leaky-bucket shaper for each flow or traffic class

59 Traffic Classification and Marking Mark a packet to influence handling downstream Early Congestion Notification (ECN) flag Type-of-Service (ToS) bits Ways to set the ToS bits End host sets the bits based on the application But, then the network must trust (or bill!) the end host Network sets the bits based on traffic classes But, then the network needs to know how to classify packets Identifying traffic classes Packet classification based on the five tuple Rate limits, with separate mark for out of profile traffic 59

60 Many Boxes, But Similar Functions Router Forward on destination IP address Access control on the five tuple Link scheduling and marking Monitoring traffic Deep packet inspection Switch Forward on destination MAC address Firewall Access control on five tuple (and more) NAT Mapping addresses and port numbers Shaper Classify packets Shape or schedule Packet sniffer Monitoring traffic 60