friends, not foes –synthesizing existing transport strategies for...

66
Friends, not Foes – Synthesizing Existing Transport Strategies for Data Center Networks Ali Munir Michigan State University Michigan State University Ghufran Baig, Syed M. Irteza, Ihsan A. Qazi, Alex X. Liu, Fahad R. Dogar

Upload: others

Post on 20-Jun-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Friends, not Foes – Synthesizing Existing Transport Strategies for Data Center Networks

Ali MunirMichigan State UniversityMichigan State University

Ghufran Baig, Syed M. Irteza, Ihsan A. Qazi, Alex X. Liu, Fahad R. Dogar

Page 2: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Data Center (DC) Applications

Distributed applicationsComponents interact via the network e.g., a bing search query touches > 100 machines

Search Mail Map-Reduce

Map-Reduce

HPC Monitoring

Network impacts performance “10% of search responses observe 1 to 14 ms of network queuing delay” [ DCTCP, SIGCOMM 10]

Image source: http://cdn.slashgear.com/wp-content/uploads/2012/10/google-datacenter-tech-13.jpg

Page 3: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Network Resource Allocation

Fair Sharing Equal bandwidth sharing among jobs [TCP, DCTCP]

– Increases completion time for everyone

– Traditional “fairness” metrics less relevant

QoS Aware Prioritize some jobs over other jobs (Priority Scheduling)

– Minimize flow completion times [pFabric, L2DCT]

– Meet flow deadlines [D3, D2TCP]

Page 4: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transports

DCTCPSIGCOMM’10

D2TCPSIGCOMM’12

L2DCTINFOCOM’13

D3

SIGCOMM’11

PDQSIGCOMM’12

pFabricSIGCOMM’13

Page 5: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transports

DCTCPSIGCOMM’10

D2TCPSIGCOMM’12

L2DCTINFOCOM’13

D3

SIGCOMM’11

PDQSIGCOMM’12

pFabricSIGCOMM’13Near Optimal but not

Deployment Friendly(Changes in data plane)

Page 6: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transports

DCTCPSIGCOMM’10

D2TCPSIGCOMM’12

L2DCTINFOCOM’13

D3

SIGCOMM’11

PDQSIGCOMM’12

pFabricSIGCOMM’13

Deployment Friendly but Suboptimal

Near Optimal but not Deployment Friendly(Changes in data plane)

Page 7: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transports

DCTCPSIGCOMM’10

D2TCPSIGCOMM’12

L2DCTINFOCOM’13

D3

SIGCOMM’11

PDQSIGCOMM’12

pFabricSIGCOMM’13

Deployment Friendly but Suboptimal

Near Optimal but not Deployment Friendly(Changes in data plane)

Step back and ask

How can we design a deployment friendly and near optimal data center transport while leveraging the

insights offered by existing proposals?

Page 8: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transports

DCTCPSIGCOMM’10

D2TCPSIGCOMM’12

L2DCTINFOCOM’13

D3

SIGCOMM’11

PDQSIGCOMM’12

pFabricSIGCOMM’13

Deployment Friendly but Suboptimal

Near Optimal but not Deployment Friendly(Changes in data plane)

Step back and ask

How can we design a deployment friendly and near optimal data center transport while leveraging the

insights offered by existing proposals?

PASE

Page 9: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Rest of the Talk …

DC Transport Strategies

PASE Design

Evaluation

Page 10: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Rest of the Talk …

DC Transport Strategies

PASE Design

Evaluation

Page 11: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transport Strategies

Self-adjusting endpoints

– senders make independent decisions and adjust rate by themselves

Arbitration

e.g., TCP, DCTCP, L2DCT

e.g., D3, PDQ

– a common network entity (e.g., a switch) allocates rates to each flow

In-network prioritization

– switches schedule and drop packets based on the packet priority

e.g., pFabric

Page 12: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

DC Transport Strategies

Self-adjusting endpoints

– senders make independent decisions and adjust rate by themselves

Arbitration

e.g., TCP, DCTCP, L2DCT

e.g., D3, PDQ

Existing DC transport proposals use – a common network entity (e.g., a switch) allocates rates to each flow

In-network prioritization

– switches schedule and drop packets based on the packet priority

e.g., pFabric

Existing DC transport proposals use only one of these strategies

Page 13: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Isolation

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Arbitration PDQ, D3

In-network Prioritization pFabric

Page 14: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Isolation

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3

In-network Prioritization pFabric

Page 15: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Isolation

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

o High flow switching

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3 Strict priority scheduling

o High flow switching overhead

o Hard to computeprecise rates

In-network Prioritization pFabric

Page 16: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Isolation

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

o High flow switching

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

o High flow switching

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3 Strict priority scheduling

o High flow switching overhead

o Hard to computeprecise rates

In-network Prioritization pFabric

Arbitration PDQ, D3 Strict priority scheduling

o High flow switching overhead

o Hard to computeprecise rates

In-network Prioritization pFabric

Low flow switchingoverhead

o Switch-local decisionso Limited # of priority

queues

Page 17: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

o High flow switching

Transport Strategy

Example Pros Cons

Self-Adjusting Endpoints

DCTCP, D2TCP, L2DCT

Ease of deploymentNo strict priority scheduling

o High flow switching

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3

In-network Prioritization pFabric

Arbitration PDQ, D3 Strict priority scheduling

o High flow switching overhead

o Hard to computeprecise rates

In-network Prioritization pFabric

Arbitration PDQ, D3 Strict priority scheduling

o High flow switching overhead

o Hard to computeprecise rates

In-network Prioritization pFabric

Low flow switchingoverhead

o Switch-local decisionso Limited # of priority

queues

Page 18: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

In-network Prioritization Alone

High Priority

Limited # of queuesMore # of flows (priorities)

High Priority

Low Priority

1

2

3

4

Flows

Page 19: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

In-network Prioritization Alone

High Priority

Limited # of queuesMore # of flows (priorities)

Flow MultiplexingLimited performance gains!

High Priority

Low Priority

1

2

3

4

Flows

Any static mapping mechanism degrades performance!

Page 20: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

In-network Prioritization + Arbitration

ArbitratorDynamic mapping of flows to queues

IdeaAs a flow’s turn comes, map it to the highest priority queue!

Page 21: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

In-network Prioritization + Arbitration

ArbitratorDynamic mapping of flows to queues

High Priority

IdeaAs a flow’s turn comes, map it to the highest priority queue!

High Priority

Low Priority

Flows 1

234

Time t1

Arbitrator

Page 22: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

In-network Prioritization + Arbitration

ArbitratorDynamic mapping of flows to queues

High Priority High Priority

IdeaAs a flow’s turn comes, map it to the highest priority queue!

High Priority

Low Priority

Flows 1

234

Time t1 Time t2

High Priority

Low Priority

Flows2

34

Arbitrator Arbitrator

Page 23: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Transport Strategies in Unison

In-network Prioritization + Arbitration

ArbitratorDynamic mapping of flows to queues

High Priority High Priority

IdeaAs a flow’s turn comes, map it to the highest priority queue!

Similarly, • Arbitration + Self-Adjusting EndpointsHigh Priority

Low Priority

Flows 1

234

Time t1 Time t2

High Priority

Low Priority

Flows2

34

Arbitrator Arbitrator

• Arbitration + Self-Adjusting Endpoints• Arbitration + In-network Prioritization

PASE leverages these insights in its design!

Page 24: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Rest of the Talk …

DC Transport Strategies

PASE Design

Evaluation

Page 25: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Design Principle

Each transport strategy should focus on

what it is best at doing!

Arbitrators

– Do inter-flow prioritization at coarse time-scales

Endpoints

– Probe for any spare link capacity

In-network prioritization

– Do per-packet prioritization at sub-RTT timescales

Page 26: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Overview

SenderReceiver

Arbitrator

Page 27: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Overview

SenderReceiver

Arbitrator

Arbitration: Control planeCalculate “reference rate” and “priority queue”

Page 28: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Overview

SenderReceiverFeedback

Arbitrator

Arbitration: Control planeCalculate “reference rate” and “priority queue”Self-Adjusting Endpoints: Guided rate control Use arbitrator feedback as a pivot

Page 29: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Overview

SenderReceiverFeedback

Arbitrator

Arbitration: Control planeCalculate “reference rate” and “priority queue”Self-Adjusting Endpoints: Guided rate control Use arbitrator feedback as a pivotIn-network Prioritization: Existing priority queues

Page 30: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Overview

SenderReceiverFeedback

Arbitrator

Arbitration: Control planeCalculate “reference rate” and “priority queue”Self-Adjusting Endpoints: Guided rate control Use arbitrator feedback as a pivotIn-network Prioritization: Existing priority queues

Key Components

Page 31: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration

SenderReceiver

Arbitrator

Page 32: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration

SenderReceiver

Arbitrator Arbitrator Arbitrator

Distributed Arbitration per link arbitration done in

control plane existing protocols implement

in data plane

Arbitrator

Page 33: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration

SenderReceiver

Arbitrator Arbitrator Arbitrator

Distributed Arbitration per link arbitration done in

control plane existing protocols implement

in data plane

Arbitrator Location at the end hosts (e.g., for their

own links to the switch) OR on dedicated hosts inside the

DC

Arbitrator

Page 34: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration

SenderReceiver

Feedback FeedbackFeedback

Arbitrator Arbitrator Arbitrator

Distributed Arbitration per link arbitration done in

control plane existing protocols implement

in data plane

Arbitrator Location at the end hosts (e.g., for their

own links to the switch) OR on dedicated hosts inside the

DC

Arbitrator

Page 35: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration

SenderReceiver

Feedback FeedbackSends data with min priority

Feedback

Arbitrator Arbitrator Arbitrator

Distributed Arbitration per link arbitration done in

control plane existing protocols implement

in data plane

Arbitrator Location at the end hosts (e.g., for their

own links to the switch) OR on dedicated hosts inside the

DC

priorityArbitrator

Page 36: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration – Challenges

Challenges

– Arbitration latency

– Processing overhead

– Network overhead

Page 37: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Arbitration – Challenges

Challenges

– Arbitration latency

– Processing overhead

– Network overhead

Solution: Leverage the tree-like structure of typical DC topologies

Page 38: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

Page 39: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

ToR

Aggregation

Core

ToR

Aggregation

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

SenderReceiver

Inter-Rack

Page 40: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

ToR

Aggregation

Core

ToR

Aggregation

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

SenderReceiver

Arbitration Message

Inter-Rack

Page 41: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

ToR

Aggregation

Core

ToR

Aggregation

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

SenderReceiver

Arbitration Message

Inter-Rack

Page 42: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

ToR

Aggregation

Core

ToR

Aggregation

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

SenderReceiver

Arbitration Message

Receiver Response

Inter-Rack

Page 43: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

ToR

Aggregation

Core

ToR

Aggregation

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

ToR

SenderReceiver

Arbitration Message

Receiver Response

Intra-RackInter-Rack

No external arbitrators required!

Sender Receiver

Page 44: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

ToR

Aggregation

Core

ToR

Aggregation

Bottom Up Arbitration

Leverage Tree Structure from leaves up to the root

ToR

SenderReceiver

Facilitates inter-rack optimizations (early pruning & delegation) to reduce arbitration overhead.

Arbitration Message

Receiver Response

Intra-RackInter-Rack

No external arbitrators required!

Sender Receiver

Page 45: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Early Pruning

Arbitration involves sorting flows and picking top k for

immediate schedulingAgg

Core

k k

TOR

Flows that won’t make it to top k queues should be pruned at lower levels

kk k

k

Page 46: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Early Pruning

Arbitration involves sorting flows and picking top k for

immediate schedulingAgg

Core

k kReduces Network and Processing overhead

TOR

Flows that won’t make it to top k queues should be pruned at lower levels

kk k

kReduces Network and Processing overheadFewer flows contact the higher level

arbitrators!

Page 47: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Delegation

Aggregation

Core

Key Idea: Divide a link into virtual links and delegate responsibility to child arbitrators

ToRs

Page 48: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Delegation

AlgorithmLink capacity C is split in N

virtual linksAggregation

Core

C

Link Capacity

Key Idea: Divide a link into virtual links and delegate responsibility to child arbitrators

ToRs

Page 49: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Delegation

AlgorithmLink capacity C is split in N

virtual linksAggregation

Core

C

Link Capacity

Delegated Capacities

Key Idea: Divide a link into virtual links and delegate responsibility to child arbitrators

Parent arbitrator delegates virtual link to child arbitratorToRs

a1 a2 aN

Delegated Capacities

Page 50: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Delegation

AlgorithmLink capacity C is split in N

virtual linksAggregation

Core

C

Link Capacity

Delegated Capacities

Key Idea: Divide a link into virtual links and delegate responsibility to child arbitrators

Parent arbitrator delegates virtual link to child arbitratorToRs

a1 a2 aN

Delegated Capacities

Child arbitrator does arbitration for virtual link

Page 51: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Delegation

AlgorithmLink capacity C is split in N

virtual linksAggregation

Core

C

Link Capacity

Delegated Capacities

Key Idea: Divide a link into virtual links and delegate responsibility to child arbitrators

Parent arbitrator delegates virtual link to child arbitrator

Virtual link capacity is periodically updated based on the top k flows of all child arbitrators

ToRs

a1 a2 aN

Delegated Capacities

Child arbitrator does arbitration for virtual link

Page 52: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Delegation

AlgorithmLink capacity C is split in N

virtual linksAggregation

Core

C

Link Capacity

Delegated Capacities

Key Idea: Divide a link into virtual links and delegate responsibility to child arbitrators

Reduces Arbitration LatencyParent arbitrator delegates

virtual link to child arbitrator

Virtual link capacity is periodically updated based on the top k flows of all child arbitrators

ToRs

a1 a2 aN

Delegated Capacities

Child arbitrator does arbitration for virtual link

Reduces Arbitration LatencyMake arbitration decision close to the

sources

Page 53: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Overview

SenderReceiverFeedback

Arbitrator

Arbitration: Control planeCalculate “reference rate” and “priority queue”

Self-Adjusting Endpoints: Guided rate control Use arbitrator feedback as a pivot

In-network Prioritization: Existing priority queues

Page 54: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Endhost Transport

Rate Control

Loss Recovery Mechanism

Page 55: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Endhost Transport

Rate Control

Use reference rate and priority feedback from arbitrators

– Use reference-rate as pivot, and

– Follow DCTCP control laws– Follow DCTCP control laws

Loss Recovery Mechanism

Page 56: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE Endhost Transport

Rate Control

Use reference rate and priority feedback from arbitrators

– Use reference-rate as pivot, and

– Follow DCTCP control laws– Follow DCTCP control laws

Loss Recovery Mechanism

– Packets in lower priority queues can be delayed for several RTTs

– large RTO OR small probe to avoid spurious retransmissions

Page 57: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

PASE -- Putting it Together

Sender Receiver

Feedback Feedback Feedback

Arbitrator Arbitrator Arbitrator

Efficient arbitration control plane

Simple TCP-like transport

Existing priority queues inside switches

Page 58: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Rest of the Talk …

DC Transport Strategies

PASE Design

Evaluation

Page 59: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Evaluation

Platforms

– Small scale testbed

– NS2

Workloads

– Web search (DCTCP), Data mining (VL2)

Comparison with deployment friendly

– DCTCP, D2TCP, L2DCT

Comparison with state of the art

– pFabric

Page 60: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Simulation Setup

Queue Size 250KB (per queue)

RTT 300usec

RTO 1 msec

L 40

Page 61: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Comparison with Deployment Friendly

Settings similar to D2TCP• Flow Sizes: 100-500KB• Deadlines: 5-25msec

Page 62: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Comparison with Deployment Friendly

Settings similar to D2TCP• Flow Sizes: 100-500KB• Deadlines: 5-25msec

PASE is deployment friendly yet performsBETTER than existing protocols!

Page 63: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Comparison with State of the Art

Settings• Flow Sizes: 2-98KB• Left-to-right traffic

percentile

99th

Page 64: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Comparison with State of the Art

Settings• Flow Sizes: 2-98KB• Left-to-right traffic

percentile

PASE performs comparable and does notrequire changes to data plane

99th

Page 65: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Summary

Key Strategies for Existing DC Transport– Arbitration, in-network Prioritization, Self-Adjusting End-

points

– Complimentary rather than substitutes

PASE– Combines the three strategies– Combines the three strategies

– Efficient arbitration control plane; simple TCP-like transport; leverages existing priority queues inside switches

Performance– Comparable to or better than earlier proposals that even

require changes to the network fabric

Page 66: Friends, not Foes –Synthesizing Existing Transport Strategies for …conferences.sigcomm.org/sigcomm/2014/doc/slides/117.pdf · PASE. Rest of the Talk … DC Transport Strategies

Thank you!Thank you!