+ All Categories
Home > Documents > Cloud Control with Distributed Rate Limiting

Cloud Control with Distributed Rate Limiting

Date post: 23-Feb-2016
Category:
Upload: chen
View: 43 times
Download: 0 times
Share this document with a friend
Description:
Cloud Control with Distributed Rate Limiting. Barath Raghavan , Kashi Vishwanath , Sriram Ramabhadran , Kenneth Yocum , and Alex C. Snoeren University of California, San Diego. Centralized network services. Hosting with a single physical presence However, clients are across the Internet. - PowerPoint PPT Presentation
Popular Tags:
32
Cloud Control with Distributed Rate Limiting Barath Raghavan, Kashi Vishwanath, Sriram Ramabhadran, Kenneth Yocum, and Alex C. Snoeren University of California, San Diego
Transcript
Page 1: Cloud Control with Distributed Rate Limiting

Cloud Control withDistributed Rate Limiting

Barath Raghavan, Kashi Vishwanath,Sriram Ramabhadran, Kenneth Yocum, and Alex C.

SnoerenUniversity of California, San Diego

Page 2: Cloud Control with Distributed Rate Limiting

Centralized network services

• Hosting with a single physical presence– However, clients are across the Internet

Page 3: Cloud Control with Distributed Rate Limiting

Running on a cloud• Resources and clients are across the

world• Services combine these distributed

resources

1 Gbps

Page 4: Cloud Control with Distributed Rate Limiting

Key challenge

We want to control distributed resources as if they were

centralized

Page 5: Cloud Control with Distributed Rate Limiting

Ideal: Emulate a single limiter

• Make distributed feel centralized– Packets should experience same limiter

behavior

S

S

S

D

D

D

0 ms

0 ms

0 ms

Limiters

Page 6: Cloud Control with Distributed Rate Limiting

Distributed Rate Limiting (DRL)

Achieve functionally equivalent behavior to a central limiter

GlobalRandom Drop

FlowProportional Share

Packet-level(general)

Flow-level(TCP specific)

GlobalToken Bucket1 2 3

Page 7: Cloud Control with Distributed Rate Limiting

Distributed Rate Limiting tradeoffs

Accuracy(how close to K Mbps is delivered, flow rate

fairness)+

Responsiveness(how quickly demand shifts are accommodated)

Vs.

Communication Efficiency(how much and often rate limiters must

communicate)

Page 8: Cloud Control with Distributed Rate Limiting

Limiter 1

DRL Architecture

Limiter 2

Limiter 3

Limiter 4

Gossip

GossipGossipEstimatelocal demand

Estimateintervaltimer

Set allocationGlobal

demand

Enforce limit

Packetarrival

Page 9: Cloud Control with Distributed Rate Limiting

Token Buckets

Token bucket, fill rate K Mbps

Packet

Page 10: Cloud Control with Distributed Rate Limiting

Demand info(bytes/sec)

Building a Global Token Bucket

Limiter 1 Limiter 2

Page 11: Cloud Control with Distributed Rate Limiting

Baseline experiment

Limiter 13 TCP flowsS D

Limiter 27 TCP flowsS D

Single token bucket10 TCP flows

S D

Page 12: Cloud Control with Distributed Rate Limiting

Global Token Bucket (GTB)

Single token bucket Global token bucket

7 TCP flows3 TCP flows

10 TCP flows

Problem: GTB requires near-instantaneous arrival info

(50ms estimate interval)

Page 13: Cloud Control with Distributed Rate Limiting

Global Random Drop (GRD)

5 Mbps (limit)4 Mbps (global arrival rate)

Case 1: Below global limit, forward packet

Limiters send, collect global rate info from others

Page 14: Cloud Control with Distributed Rate Limiting

Global Random Drop (GRD)

5 Mbps (limit)6 Mbps (global arrival rate)

Case 2: Above global limit, drop with probability:Excess

Global arrival rate

Same at all limiters

16=

Page 15: Cloud Control with Distributed Rate Limiting

GRD in baseline experiment

Single token bucket Global random drop

7 TCP flows3 TCP flows

10 TCP flows

(50ms estimate interval)

Delivers flow behavior similar to a central limiter

Page 16: Cloud Control with Distributed Rate Limiting

GRD with flow join(50ms estimate interval)

Flow 1 joins at limiter 1

Flow 2 joins at limiter 2Flow 3 joins at limiter 3

Page 17: Cloud Control with Distributed Rate Limiting

Flow Proportional Share (FPS)Limiter 1

3 TCP flowsS D

Limiter 27 TCP flows

S D

Page 18: Cloud Control with Distributed Rate Limiting

Flow Proportional Share (FPS)

Limiter 1 Limiter 2

“3 flows”“7 flows”

Goal: Provide inter-flow fairness for TCP flows

Local token-bucketenforcement

Page 19: Cloud Control with Distributed Rate Limiting

Estimating TCP demandLimiter 1

D

Limiter 23 TCP flows

S D

1 TCP flow

S1 TCP flow

S

Page 20: Cloud Control with Distributed Rate Limiting

Estimating TCP demand

Local token rate (limit) = 10 Mbps

Flow A = 5 Mbps

Flow B = 5 Mbps

Flow count = 2 flows

Page 21: Cloud Control with Distributed Rate Limiting

Estimating TCP demandLimiter 1

1 TCP flowS

D

Limiter 23 TCP flows

S D

S 1 TCP flow

Page 22: Cloud Control with Distributed Rate Limiting

Key insight: Use a TCP flow’s rate to infer demand

Estimating skewed TCP demand

Local token rate (limit) = 10 MbpsFlow A = 2 Mbps

Flow B = 8 Mbps

Flow count ≠ demand

Bottlenecked elsewhere

Page 23: Cloud Control with Distributed Rate Limiting

Estimating skewed TCP demand

Local token rate (limit) = 10 MbpsFlow A = 2 Mbps

Flow B = 8 Mbps

Local LimitLargest Flow’s Rate

108=

Bottlenecked elsewhere

= 1.25 flows

Page 24: Cloud Control with Distributed Rate Limiting

2.50 flowsLimiter 2

10 Mbps x 1.251.25 + 2.50

Flow Proportional Share (FPS)Global limit = 10 Mbps

1.25 flowsLimiter 1

Set local token rate =

= 3.33 Mbps

Global limit x local flow countTotal flow count

=

Page 25: Cloud Control with Distributed Rate Limiting

Under-utilized limitersLimiter 1

DS 1 TCP flowS 1 TCP flow

Set local limit equal to actual usage

Wasted rate

(limiter returns to full utilization)

Page 26: Cloud Control with Distributed Rate Limiting

Flow Proportional Share (FPS)(500ms estimate interval)

Page 27: Cloud Control with Distributed Rate Limiting

Additional issues• What if a limiter has no flows and

one arrives?• What about bottlenecked traffic?• What about varied RTT flows?• What about short-lived vs. long-lived

flows?

• Experimental evaluation in the paper– Evaluated on a testbed and over

Planetlab

Page 28: Cloud Control with Distributed Rate Limiting

Cloud control on Planetlab• Apache Web servers on 10 Planetlab

nodes• 5 Mbps aggregate limit• Shift load over time from 10 nodes to

4 nodes

5 Mbps

Page 29: Cloud Control with Distributed Rate Limiting

Static rate limitingDemands at 10 apache servers on Planetlab

Demand shifts to just 4 nodesWasted capacity

Page 30: Cloud Control with Distributed Rate Limiting

FPS (top) vs. Static limiting (bottom)

Page 31: Cloud Control with Distributed Rate Limiting

Conclusions• Protocol agnostic limiting (extra cost)

– Requires shorter estimate intervals• Fine-grained packet arrival info not

required– For TCP, flow-level granularity is

sufficient• Many avenues left to explore

– Inter-service limits, other resources (e.g. CPU)

Page 32: Cloud Control with Distributed Rate Limiting

Questions!


Recommended