Skip to content

9.1programming_SDNs.md

Latest commit

 

History

History
138 lines (120 loc) · 4.29 KB

9.1programming_SDNs.md

File metadata and controls

138 lines (120 loc) · 4.29 KB

Consistent Updates in SDNs

Consistency Problems

  • Paket-level: Updates may disrupt packets along an end-to-end path
  • Flow-level: Packets from the same flow may be disrupted due to mid-flow update

3 Steps in SDN programming

  1. Read/monitor network states
  • failures, topology changes, security events, ...
  1. Compute Policies based on states
  • decision plane
  1. Write Policies

Reading state problems

  • Example: Web server traffic except source 1.2.3.4
    • Solution: Predicates
      • (srcip !== 1.2.3.4) && (srcport == 80)
    • Runtime system translates predicates to low-level openflow rules
  • Problem: switches have limite storage space for rules
    • cannot install all possible patterns
    • Solution: Dynamically "unfold" rules as traffic arrives
      • Programmer specify "GroupBy(srcip)"
      • Runtime system dynamically adds rules as traffic arrives
  • Problem: Extra Unexpected Events
    • 1st packet goes to controller and controller caches rules on switch
    • What if more packets arrive before rule is installed?
      • Programmer specifies "Limit(1)"
      • Runtime system hides any packets over 1

Writing Policy Problems

  • Maintenance

  • Unexpected failure

  • Traffic Engineering

  • Example: traffic engineering

    • Traffic Engineering
    • Solution: Two-Phase Commit
      • packets are tagged with version numbers
      • policies correspond to packet version numbers and old policy is kept until all packets with P1 are gone

Network Virtualization

  • Abstraction of physical network
    • Multiple logical networks on a shared physical substrate
    • Nodes in physical topology are VMs
    • Links in physical network sliced into tunnels
    • VM Analogy
  • Why Virtual Networking?
    • "Ossification" of Internet architecture.
    • enables evolution by letting multiple architecture exist in parallel
    • Practice: multi-tenant data centers
      • EC2, Rackspace, Google App engines
  • Benefits of Network Virtualization
    • Rapid innovation
    • new forms of network control
    • (potentially) simpler programming

Design Goals for VN

  • Flexible

  • Manageable

  • Scalable

  • Secure

  • Programmable

  • Able to support different technologies

  • Implementation

  • Nodes

    • VMs or Virtual Environments
    • examples: Xen, VMWare, Vservers

  • Edges

    • Tunnels
    • Simulate a layer 2 connection even over multiple layer 3 hops
    • encrypt packets on src node and decrypt on dest. node

  • Summary

    • VNs facilitate flexible, agile deployment
      • Rapid innovation
      • Vendor independence
      • Scale
    • SDNs vs. VNs
    • Technologies: VMs, Tuneling

Programming SDNs: Why & How?

  • Problem: OpenFlow not easy!
    • Low level of abstraction
    • Controller only sees events that switches do not know how to handle
    • race conditions if switch-level rules not installed properly
  • Solution: "Northbound" API
    • Higher level API so controller can talk to application & vice versa
    • Vendor independence
    • Exaples:
      • large virtual switch
      • security apps
      • middlebox integration

Frenetic: SQL-Like Query Language

  • Example:

    • select (bytes) where (in: 2 & srcport: 80) groupBy(dstMAC) every(60).

  • Problem with writing at high level of abstraction:

    • Modules affect same traffic
    • Monitors, routes, firewall rules, load balancing applications
    • Composition operators - combine modules to single, coherent, openflow policies

  • Policy Composition

    • Parallel: Perform both operations simultaneously
      • counting and forwarding
    • Sequential: Perform one operation, then next
      • firewall rules, then packets that make thru firewall are switched

Pyretic

  • SDN Language and Runtime

  • Language: express policies

  • Runtime: Compliling these policies to OpenFlow

  • Key abstraction: "located" packets

  • Network policy as function

  • Boolean predicates

  • Virtual packet header fields

  • Composition operators

  • functions

  • Policy Composition

    • Sequential Composition
      • match(dstIP=2.2.2.8)>>fwd(1)
    • Parallel Composition
      • match(dstIP=2.2.2.8)>>fwd(1) + match(dstIP=2.2.2.9)>>fwd(2)

  • Traffic Monitoring

    • Query on packet streams
      • self.query = packets(1,['scrMAC','switch'])
      • self.query.register.callback(learn_new_mac)

  • Dynamic Policies

    • Timeseries of static policies
      • current value: self.policy