Protocol Independence. Nick McKeown Stanford University

Transcription

1 Protocol Independence Nick McKeown Stanford University

2 Where did SDN ideas start? Stanford gets too much credit Roots are in 4D and RCP Rexford, Greenberg, Zhang, Maltz, SANE/Ethane from Stanford + Berkeley Many concepts from Nicira Network OS: NOX OpenFlow Distributed OS: ONIX New concepts all the Tme Berkeley, Berlin, Cornell, Gatech Princeton, Stanford, Urbana,

3 Is SDN any network device with behavior defined in sovware?

4 Is SDN anything with an OpenFlow interface?

5 Specialized Features Specialized Control Plane Specialized Hardware Hundreds of protocols 6,500 RFCs Tens of millions of lines of code. Closed, proprietary, outdated. Billions of gates. Power hungry and bloated.

6 What SDN isn t Ram in even more lines of code My box now has an OpenFlow interface too!

7 What SDN isn t Management Software Old Control Plane Custom Hardware Old Control Plane Custom Hardware Old Control Plane Custom Hardware

8 A network in which the control plane is physically separate from the forwarding plane. and A single control plane controls several forwarding devices. (That s it)

9 It s just an idea and a startng point.

10 Technical Consequences Makes crystal clear the distributed systems problem. Tells us to solve the distributed systems problem once, rather than multple Tmes. Makes it easier to control diverse switches. ParTcularly if they are protocol independent.

11 Where is OpenFlow headed?

12 OpenFlow Goals 1. Protocol- independent forwarding. 2. That can be controlled and repurposed in the field by a remote control plane. 3. [And can be implemented by really fast, low power, switching chips]

13 Currently OpenFlow v1.x It is protocol dependent because 1. Constraints of traditonal switching chips. - Each table Ted to a specific protocol - Fixed sequence processing pipeline 2. Maps to existng switching chips. - For quick adopton

14 Match- AcTon Forwarding AbstracTon Ac#on Primi#ves 1. Forward to ports 4 & 5 2. Push header Y aver bit Pop header bits Decrement bits Drop packet 6. H H Match Action F Action(F) G H Action(G) Action(H)

15 MulTple Table Match- AcTon H n H 1 H Match Action Match Action F 1 Action(F) F n Action(F) G 1 Action(G) G n Action(G) H 1 Action(H) H n Action(H)

16 What would an OpenFlow- optmized switching chip look like? Parse any existng header Parse any custom header Match: Huge protocol independent tables (Millions of entries) Many tables in a pipeline (8 or more) Tables can be used efficiently AcTon: Protocol independent InstrucTon primitves for processing headers

17 Design Exercise with Texas Instruments 64 x 10GE OpenFlow- optmized ASIC Industry- standard 28nm design process Parse existng + custom packet headers 32 stages of Match + AcTon Large tables 1M x 40b TCAM 370Mb SRAM (hash table & statstcs counters) VLIW acton processing

18 Taken as given Fast interface to local Usual features: buffers, counters, etc. Open interface, not hidden behind NDA.

19 QuesTon How much extra area and power compared to a traditonal, fixed functon switch chip?

20 RISC- like architecture Stage 1 Stage 32 HEADER Output Match Match Queues Table Table Data In H Data H AcTon AcTon DATA Recombine Out

21 Match Tables TCAM Physical Stage 1 Logical Table 1 Physical Stage 2 Logical Table 2 Physical Stage 32 Logical Table HASH 6 7 Logical Table 8

22 Physical Stage 1 Physical Stage 2 Physical Stage 32 TCAM In HEADER HASH Match Table AcTon Match Table AcTon Out DATA Recombine Output Queues

23 Stage 1 Stage 32 In HEADER Match Table AcTon Match Table AcTon Recombine Out DATA Output Queues

24 AcTon Processor Header In Field Field Header Out Match result VLIW InstrucTons

25 100s of VLIW s Match result VLIW InstrucTons

26 Stage 1 Stage 32 In HEADER Match Table AcTon Match Table AcTon Recombine Out DATA Output Queues

27 In HEADER Match Table VLIW InstrucTons AcTon Match Table VLIW InstrucTons AcTon Out DATA Recombine Output Queues

28 Takeaways Extreme flexibility Custom packet formats Repurpose in the field Large TCAM no longer a problem 1M+ entries Used very efficiently All for < 15% extra area

29 What does this tell us about OpenFlow?

30 What is possible Protocol- independent processing Pipeline: Parse + MulTple Match- AcTon Stages Field configurable pipeline

31 But wait. Won t OpenFlow commoditze my business?

32 Codeword for: Is it going to erode my enormous profit margins?

33 CompeTTon based on 1. Switching capacity, power, price speed 2. DifferenTaTng features in additon to OpenFlow Base instructon set Enabling great sovware.

34 business is healthy, profitable, innovatve. The switch chip business can be too.

35 <end>