Enabling ECN over Generic

Enabling ECN over Generic Enabling ECN over Generic - Start

Added : 2017-03-27 Views :38K

Download Presentation

Enabling ECN over Generic




Download Presentation - The PPT/PDF document "Enabling ECN over Generic" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.



Presentations text content in Enabling ECN over Generic

Slide1

Enabling ECN over Generic

Packet Scheduling

Wei Bai, Kai Chen, Li Chen, Changhoon Kim, Haitao Wu

ACM CoNEXT, Irvine, CA, December 2016

1

Slide2

Data Centers Around the World

Google’s worldwide DC map

2

Microsoft’s DC in

Dublin, Ireland

Facebook DC interior

Global Microsoft Azure DC Footprint

ACM CoNEXT, Irvine, CA, December 2016

Slide3

Inside the Data Center (DC)

Network requirements of applicationsDesire low latency for short messagesDesire high throughput for large flows

ACM CoNEXT, Irvine, CA, December 2016

3

Slide4

Inside the Data Center (DC)

Network requirements of applicationsDesire low latency for short messagesDesire high throughput for large flowsNetwork performance improvementPacket schedulingECN-based transport protocols

ECN = Explicit Congestion Notification

ACM CoNEXT, Irvine, CA, December 2016

4

Combine

Slide5

Packet Scheduling in Data Centers

5

Round Robin

Real-time Services

Best-effort Services

Background Services

4

2

1

Weight

Inter-Service Traffic Isolation

Bai et al. (NSDI’16)

ACM CoNEXT, Irvine, CA, December 2016

Slide6

Packet Scheduling in Data Centers

6

Round Robin

(0, 100KB] Flows

(100KB, 10MB) Flows

(10MB,

) Flows

 

High

Medium

Low

Priority

Flow Scheduling

Bai et al. (NSDI’15)

Strict Priority

ACM CoNEXT, Irvine, CA, December 2016

Slide7

Packet Scheduling in Data Centers

7

Round Robin

Existing fixed-function

switching chips

Strict Priority

ACM CoNEXT, Irvine, CA, December 2016

Slide8

Packet Scheduling in Data Centers

8

Round Robin

Strict Priority

Future programmable

switching chips

ACM CoNEXT, Irvine, CA, December 2016

Slide9

Packet Scheduling in Data Centers

9

Round Robin

Programmable

Schedulers

Push-In-First-Out (PIFO) Queue

A

Sivaraman

et al. (SIGCOMM’16)

Strict Priority

ACM CoNEXT, Irvine, CA, December 2016

Slide10

Can we enable ECN for arbitrary packet schedulers in data centers?

10

ACM CoNEXT, Irvine, CA, December 2016

Slide11

Packets get marked when queue length

 

ACM CoNEXT, Irvine, CA, December 2016

11

 

don’t mark

mark

ECN/RED without Packet Scheduling

Slide12

ECN/RED without Packet Scheduling

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

12

Buffer Occupancy

K

Time

Slide13

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

13

 

Small number of concurrent large flows in DC

M

Alizadeh et al. (SIGCOMM’10)

ECN/RED without Packet Scheduling

Slide14

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

14

 

Fixed link capacity

ECN/RED without Packet Scheduling

Slide15

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

15

 

Base round-trip time, relatively stable in DC

Wu et al. (CoNEXT’12)

ECN/RED without Packet Scheduling

Slide16

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

16

 

Determined by congestion control algorithms

ECN/RED without Packet Scheduling

Slide17

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

17

 

Standard queue length threshold

A static value in data center environment

ECN/RED without Packet Scheduling

Slide18

Packets get marked when queue length To achieve 100% throughput

 

ACM CoNEXT, Irvine, CA, December 2016

18

static

threshold:

 

Easy to configure at the switch

ECN/RED without Packet Scheduling

Slide19

ECN/RED with Packet Scheduling

Each queue is a link with the varying capacityIdeal ECN/RED solutionPackets should get marked if the length of queue i

 

ACM CoNEXT, Irvine, CA, December 2016

19

dynamic

per-queue threshold:

 

varying

capacity:

 

Slide20

ECN/RED with Packet Scheduling

Each queue is a link with the varying capacityIdeal ECN/RED solutionPackets should get marked if the length of queue i Not supported by current switching chipsCurrent practiceConfigure static thresholds: High throughput but poor latency

 

ACM CoNEXT, Irvine, CA, December 2016

20

Slide21

To Implement Ideal ECN/RED Solution

A general way to estimate the queue capacityQueue capacity = Queue departure rate when the queue keeps non-emptyLeverage the solution from PIE (HPSR’13)Start measurement when # of bytes in the switch buffer > dq_threshGet the rate to drain dq_thresh bytes

ACM CoNEXT, Irvine, CA, December 2016

21

Slide22

Trade-off of Measurement Window

ACM CoNEXT, Irvine, CA, December 2016

22

Sequence of packets

Link capacity: C

Transmitted packets from queue 1

Transmitted packets from queue 2

Queue 1 and 2 keep

non-empty

during the transmission

Slide23

Trade-off of Measurement Window

ACM CoNEXT, Irvine, CA, December 2016

23

Sequence of packets

Link capacity: C

Transmitted packets from queue 1

Transmitted packets from queue 2

Queue capacity 1 = Queue capacity 2 = 0.5C

Slide24

Trade-off of Measurement Window

A too small measurement windowe.g., dq_thresh = 3MTU

ACM CoNEXT, Irvine, CA, December 2016

24

Sequence of packets

Link capacity: C

C

3/7 C

3/7 C

C

Sample rate of queue 1

Slide25

Trade-off of Measurement Window

A too small measurement windowDegrade measurement accuracy

ACM CoNEXT, Irvine, CA, December 2016

25

Sequence of packets

Link capacity: C

C

3/7 C

3/7 C

C

Sample rate of queue 1

Slide26

Trade-off of Measurement Window

A too small measurement windowDegrade measurement accuracyA too large measurement windowe.g, dq_thresh = 20MTU

ACM CoNEXT, Irvine, CA, December 2016

26

Sequence of packets

Link capacity: C

Slide27

Trade-off of Measurement Window

A too small measurement windowDegrade measurement accuracyA too large measurement windowCannot efficiently capture the dynamic changes

ACM CoNEXT, Irvine, CA, December 2016

27

Sequence of packets

Link capacity: C

Slide28

Trade-off of Measurement Window

A too small measurement windowDegrade measurement accuracyA too large measurement windowCannot efficiently capture the dynamic changes

ACM CoNEXT, Irvine, CA, December 2016

28

Rate measurement is non-trivial

Slide29

Another View

Ideal ECN/RED solutionPackets should get marked if

 

ACM CoNEXT, Irvine, CA, December 2016

29

varying

capacity:

 

queue length:

 

Slide30

Another View

Ideal ECN/RED solutionPackets should get marked if

 

ACM CoNEXT, Irvine, CA, December 2016

30

varying

capacity:

 

sojourn time:

 

Slide31

TCN

TCN mechanismPackets should get marked if their sojourn times >

 

ACM CoNEXT, Irvine, CA, December 2016

31

T

ime-based

C

ongestion

N

otification

Slide32

TCN in Detail

Sojourn time measurementEnqueue: attach a metadata to each packet to store the enqueue time

ACM CoNEXT, Irvine, CA, December 2016

32

T

eq

Slide33

TCN in Detail

Sojourn time measurementEnqueue: attach a metadata to each packet to store the enqueue timeDequeue: calculate sojourn time

ACM CoNEXT, Irvine, CA, December 2016

33

T

eq

sojourn time = now -

T

eq

Slide34

TCN in Detail

Sojourn time measurementEnqueue: attach a metadata to each packet to store the enqueue timeDequeue: calculate sojourn time

ACM CoNEXT, Irvine, CA, December 2016

34

2B-long

metadata is enough for DC

Slide35

TCN in Detail

Sojourn time measurementEnqueue: attach a metadata to each packet to store the enqueue timeDequeue: calculate sojourn timeInstantaneous ECN markingCompare the per-packet instantaneous sojourn time with a static threshold

 

ACM CoNEXT, Irvine, CA, December 2016

35

Stateless

Data Plane Algorithm

Slide36

TCN in Detail

Sojourn time measurementEnqueue: attach a metadata to each packet to store the enqueue timeDequeue: calculate sojourn timeInstantaneous ECN markingCompare the per-packet instantaneous sojourn time with a static threshold Marking does not cause any bubble on the link

 

ACM CoNEXT, Irvine, CA, December 2016

36

Slide37

TCN vs CoDel

Advantages of TCNStateless: cheaper to implement in hardwareInstantaneous: faster reaction to busty traffic

ACM CoNEXT, Irvine, CA, December 2016

37

Slide38

TCN vs CoDel

Advantages of TCNStateless: cheaper to implement in hardwareInstantaneous: faster reaction to busty trafficSimplicity of TCN

ACM CoNEXT, Irvine, CA, December 2016

38

Unique Characteristics of Data Centers

Slide39

TCN vs CoDel

Advantages of TCNStateless: cheaper to implement in hardwareInstantaneous: faster reaction to busty trafficSimplicity of TCNSmall number of concurrent large flows Relatively stable RTTsPrior knowledge of transport at the end host

ACM CoNEXT, Irvine, CA, December 2016

39

Slide40

Testbed Evaluation

TCN software prototypeLinux qdisc kernel module on a multi-NIC serverTestbed setup9 servers are connected to a software switchEnd-hosts use DCTCP as the transport protocolECN schemes comparedPer-queue RED with the standard thresholdCoDel

40

Slide41

Static Flow Experiment

1 flow (500Mbps)

4 flows

41

1 flow

Q1

Q2

Q3

SP/WFQ

w=1 (low)

w=1 (low)

high prio

Slide42

Static Flow Experiment

ACM CoNEXT, Irvine, CA, December 2016

42

TCN preserves the scheduling policy

Slide43

Dynamic Flow Experiment

43

high prio

w=1 (low)

w=1 (low)

w=1 (low)

w=1 (low)

8 senders to 1 receiver (web search workload)

SP/WFQ scheduling policy at the switch

SP/WFQ

(0, 100KB] flows of all services

(100KB,

) flows of service 1

 

(100KB,

) flows of service 2

 

(100KB,

) flows of service 3

 

(100KB,

) flows of service 4

 

Traffic

Slide44

ACM CoNEXT, Irvine, CA, December 2016

44

99th FCT of Small Flows (<100KB)

TCN maintains the low buffer occupancy

Slide45

ACM CoNEXT, Irvine, CA, December 2016

45

Realistic Traffic: Large Flows (>10MB)

TCN achieves high throughput

Slide46

Conclusion

TCN: a simple ECN solution for data centersUse sojourn time as the congestion signal (CoDel)Perform instantaneous ECN marking (DCTCP)Code: http://sing.cse.ust.hk/projects/TCNNext step: TCN in programmable hardware

ACM CoNEXT, Irvine, CA, December 2016

46

Slide47

Thanks!

47

Slide48

ACM CoNEXT, Irvine, CA, December 2016

48

Average FCT of Small Flows (<100KB)


About DocSlides
DocSlides allows users to easily upload and share presentations, PDF documents, and images.Share your documents with the world , watch,share and upload any time you want. How can you benefit from using DocSlides? DocSlides consists documents from individuals and organizations on topics ranging from technology and business to travel, health, and education. Find and search for what interests you, and learn from people and more. You can also download DocSlides to read or reference later.
Youtube