Catalyst: Unlocking Power of Choice to Speed up Network Updates - - PowerPoint PPT Presentation

Catalyst: Unlocking Power of Choice to Speed up Network Updates

Rohan Gandhi, Ori Rottenstreich, Xin Jin

1

Network Update Cycle

Network Controller Network Updater

Target state Compute network target state Update network to target state

2

Current state (Traffic, topology etc.) Target state (Rules) Centralized Controller

Network Update Requirements

Consistent

• No congestion
• No traffic blackhole
• No loop

Fast

• Reduce failure impact
• Improve network optimality

3

Dependency Graph in Existing Network Updaters

4 S1 S3 S4

f1:4 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6

S1 S3 S4

f1:4 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6 Consistent network update plan Mv f1 Mv f2 Mv f4

Start

Initial State Target State

Naive (one-shot) update plan (Can cause inconsistencies) Mv f1, f4, f2

Start

Link capacity = 10 units B4 (SIGCOMM 2013): Network update takes 3-5X more time than controller

Existing network updaters assume target state cannot be changed

5

Limitation 1: Long Update Plans

S1 S3 S4

f1:4 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6

S1 S3 S4

f1:4 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6 Update plan Mv f1 Mv f2 Mv f4

Start 6

Initial State Target State The controller has no information about the length (longest path)

• f the dependency graph

Limitation 2: Inability to Effectively Tackle Stragglers

7

Straggling switches can take 10x or more time to update rules Even a single update through a straggling switch can delay entire stage in dependency graph

Dionysus, SIGCOMM 2014

Catalyst

• Redundancy in the network offers power of choice
• Use power of choice to change the target state and

speed up network update

8

Key idea 1: Shorten Dependency Graph

9 S1 S3 S4

f1:4 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6

S1 S3 S4

f1:4 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6 Update plan Mv f1 Mv f2 Mv f4

Start

Initial State Initial Target State

S1 S3 S4

f1:5 f4:6 f2:5

S2

f5:5

S5

f3:4 f6:6 Mv f1 Mv f2 Mv f4 Star

Modified Target State (+) Merging stages reduces number of stages, which reduces update time (-) Merging stages increases number of updates in a stage, which increases probability

• f a straggler

Merging stages always reduces update time (proof in paper)

Key idea 2: Multiple Paths to Tackle Stragglers

10

Assign single flow to multiple (equally optimal) paths

S1 S2 S3 S4 S6 Straggler Ingress Egress S2 Update time Catalyst Default S5 S3 S2 S3 S5 S4

Try all paths for a flow in parallel

Key Challenges

• Which flows to move to alternate paths?
• How many alternate paths to choose?
• How to compute alternate paths?
• How to schedule alternate paths?

11

Evaluation Setup

• Load-balancer in datacenter (similar to “Incremental Consistent Updates”, HotSDN 2013)
• Datacenter topology:
• 100 ToRs in 50 containers with 10Gbps links
• Number of flows = 10K
• Load balancer settings:
• Assigns flows to replicas uniformly
• Initial number of servers = 100
• Update settings:
• We fail 2 servers and reassign flows to remaining replicas

12

Evaluation

13

“s” = fraction of redundancy compared to total capacity

• (s=0.1) Median improvement in network update latency
• (max. Path = 1), improvement = 1.22x
• (max. Path = 2), improvement = 1.43x
• (s=0.2) Median improvement in network update latency
• (max. Path = 1), improvement = 1.32x
• (max. Path = 2), improvement = 1.65x

Increasing number of paths per flow: (+) more alternatives to tackle straggler (-) less number of flows assigned to alternate paths

Conclusion

• Existing network update solutions assume inflexibility in changing target state
• Limitation 1: Unnecessarily long network update dependency graphs
• Limitation 2: Cannot handle stragglers effectively
• Catalyst: Speeds up network updates by exploiting redundancy in the network
• Merges states in the dependency graph using redundant paths
• Assigns redundant paths to individual flows to tackle stragglers
• Evaluation using load balancer settings shows speed-up up to 1.65x.

14