KULFI. Robust Traffic Engineering Using Semi- Oblivious Routing TASTES GREAT, NO CHURN!

Transcription

1 TASTES GREAT, NO CHURN! KULFI Robust Traffic Engineering Using Semi- Oblivious Routing Praveen Kumar, Yang Yuan, Chris Yu, Bobby Kleinberg, Robert Soulé, & Nate Foster Cornell, Carnegie Mellon, Microsoft Research, & USI Lugano

2 2 WIDE-AREA TRAFFIC ENGINEERING CHALLENGES

3 2 WIDE-AREA TRAFFIC ENGINEERING CHALLENGES Unstructured topology

4 2 WIDE-AREA TRAFFIC ENGINEERING CHALLENGES Unstructured topology Unexpected Failures

5 2 WIDE-AREA TRAFFIC ENGINEERING CHALLENGES Unstructured topology Misprediction & Bursts Unexpected Failures

6 2 WIDE-AREA TRAFFIC ENGINEERING CHALLENGES Unstructured topology Misprediction & Bursts Unexpected Failures Device Limitations

7 3 ROUTING SCHEME

8 3 ROUTING SCHEME 1. Which forwarding paths to use send traffic from sources to destinations?

9 3 ROUTING SCHEME 1/2 1. Which forwarding paths to use send traffic from sources to destinations? 2. How to map incoming traffic flows onto multiple forwarding paths? 1/2

10 4 ECMP 1. Pre-compute a set of least-cost paths 2. Identify flows by hashing packet header fields 3. Randomly forward along least cost paths

11 4 ECMP 1. Pre-compute a set of least-cost paths 2. Identify flows by hashing packet header fields 3. Randomly forward along least cost paths

12 5 VALIANT LOAD BALANCING

13 5 VALIANT LOAD BALANCING I I I

14 6 VALIANT LOAD BALANCING I I I

15 6 VALIANT LOAD BALANCING I I I

16 7 OBLIVIOUS ROUTING (RÄCKE 08) Idea: if we can somehow limit ourselves to shorter paths can we get lowcongestion routing schemes? A randomized routing tree is probability distribution over routing trees We need a way to construct routing trees that have low stretch...

17 8 RÄCKE S ALGORITHM + SEMI-OBLIVIOUS ROUTING Iteratively constructs a randomized routing tree At each iteration, it penalizes edges that have been heavily utilized in previous trees Achieves a poly-logarithmic competitive ratio with respect to optimal regardless of the demand matrix i.e. it is oblivious! Semi-oblivious routing combines Räcke s oblivious routing with dynamic rate adaptation and local failure recovery Compute forwarding paths statically Adjust sending rates dynamically Performance of semi-oblivious routing in practice not known...

18 9 IMPLEMENTATION AND EVALUATION * HARDWARE TESTBED * SIMULATION

19 10 SDN IMPLEMENTATION Linux End Host Linux Kernel User- Space Agent Netfilter Module Traffic matrix + Path map forwarding rules data traffic SDN Controller SDN Switch Historical Data Traffic statistics Traffic statistics

20 11 ABILENE TOPOLOGY Emulated Abilene topology in hardware test bed S10 S8 S11 S4 S5 S7 S3 S6 S2 S1 S12 S9 Used real-world and worst case traffic scenarios Compared shortest-path, ECMP, MCF, oblivious, and semi-oblivious

21 12 ABILENE TOPOLOGY: SIMULATED WORKLOAD 1 Abilene Gravity Traffic Link congestion ECMP Time (minutes) SPF max ECMP max Obliv max Semi Obliv max MCF max SPF median ECMP median Obliv median Semi Obliv median MCF median

22 12 ABILENE TOPOLOGY: SIMULATED WORKLOAD 1 Abilene Gravity Traffic Link congestion ECMP 0 Overlap SPF max ECMP max Obliv max Semi Obliv max MCF max Time (minutes) SPF median ECMP median Obliv median Semi Obliv median MCF median

23 13 ABILENE TOPOLOGY: SIMULATED WORKLOAD S11 S10 S4 S7 S3 S6 S12 S9 S8 S5 S2 S1

24 13 ABILENE TOPOLOGY: SIMULATED WORKLOAD S11 S10 Artificial traffic S4 S7 S3 S6 S12 S9 S8 S2 S5 S1

25 14 ABILENE TOPOLOGY: SIMULATED WORKLOAD 1 Abilene Gravity + Artificial Traffic Link congestion ECMP Time (minutes) SPF max ECMP max Obliv max Semi Obliv max MCF max SPF median ECMP median Obliv median Semi Obliv median MCF median

26 14 ABILENE TOPOLOGY: SIMULATED WORKLOAD 1 Abilene Gravity + Artificial Traffic Link congestion ECMP 0 Overlap SPF max ECMP max Obliv max Semi Obliv max MCF max Time (minutes) SPF median ECMP median Obliv median Semi Obliv median MCF median

27 15 KULFI FRAMEWORK Implemented over a dozen different traffic engineering schemes Measure performance in simulator and hardware testbed with a variety of demands and failures Used local failure recovery

28 15 KULFI FRAMEWORK Implemented over a dozen different traffic engineering schemes SWAN Measure performance in simulator and hardware testbed with a variety of demands and failures Used local failure recovery

29 15 KULFI FRAMEWORK Implemented over a dozen different traffic engineering schemes SWAN Measure performance in simulator and hardware testbed with a variety of demands and failures Used local failure recovery

30 15 KULFI FRAMEWORK Implemented over a dozen different traffic engineering schemes SWAN Measure performance in simulator and hardware testbed with a variety of demands and failures Used local failure recovery Suchara et al SIGMETRICS 11

31 16 TOPOLOGY ZOO: FAULT TOLERANCE % Loss due to Failure Time

32 16 TOPOLOGY ZOO: FAULT TOLERANCE % Loss due to Failure Time

33 17 TOPOLOGY ZOO: LATENCY JANET Geant Fraction Delivered Latency (Path Length)

34 17 TOPOLOGY ZOO: LATENCY JANET Geant Fraction Delivered Latency (Path Length)

35 17 TOPOLOGY ZOO: LATENCY JANET Geant Fraction Delivered Latency (Path Length)

36 18 CONCLUSIONS Randomization can dramatically simplify traffic engineering while balancing competing objectives Oblivious routing performs much better in practice than expected, avoids problems associated with churn, and load-balances better Semi-oblivious routing provides near-optimal performance in real-world scenarios, even in the presence of demand mis-prediction, traffic bursts, and failures Ongoing work: working with large ISP and content provider to further refine and evaluate Kulfi