CA-DAG: Communication-Aware Directed Acyclic Graphs for Modeling Cloud Computing Applications...

CA-DAG: Communication-Aware Directed Acyclic Graphs for Modeling Cloud Computing

Applications

Dzmitry Kliazovich University of Luxembourg, LuxembourgJohnatan E. PeceroPascal Bouvry

Andrei Tchernykh CICESE Research Center, Mexico

Samee U. Khan North Dakota State University, U.S.A.

Albert Y. Zomaya University of Sydney, Australia

Dzmitry Kliazovich ([email protected]) 2

Cloud Computing Applications

Jul 1, 2013


Resource Requirements of Cloud Applications

Jul 1, 2013

Computing Network Bandwidth

Communication delays (tolerance)

Degree of interactivity Storage


Resource Requirements of Cloud Applications

Jul 1, 2013

Computing Network Bandwidth

Communication delays

(tolerance)StorageDegree of

interactivity



Jul 1, 2013

Communicationresources



• Traditional resource allocation and scheduling– Distribute incoming jobs to the pool of servers– Communication requirements and networking are not

taken into account

Jul 1, 2013


Scheduling in Data Centers

Jul 1, 2013

S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S

CoreNetwork

AggregationNetwork

AccessNetwork

S

Links

10 GE 1 GE

Nodes

L3 Switch L2/L3 Rack Switch Computing Server

Network congestion!!!


Scheduling in Data Centers

Jul 1, 2013


CoreNetwork

AggregationNetwork

AccessNetwork

S

Links

10 GE 1 GE

Nodes


Network is balanced !!!

Modeling Cloud Computing Applications


Modeling of Cloud Applications

• Directed Acyclic Graphs (DAGs)– Vertices represent computing tasks of a job– Edges represent task dependencies and order

of execution

Jul 1, 2013

1

2

3 4

How to model communication processes?



• Communication-unaware model

• Edges-based model

Jul 1, 2013



• Communication-unaware model– Each vertex represents both computing and

communication processes of a task

Jul 1, 2013

1

32

4Communication job

of a task

Computing jobof a task

• Main drawback– Having a single vertex for both computing and communications

makes it impossible to make separate scheduling decisions



• Edge-based model– DAG edges represent communication processes of a task

Jul 1, 2013

1

32

4

Communication jobof a task

Computing jobof a task

• Main drawback– Two different computing tasks cannot have the same data

transfer to receive input as a singe edge cannot lead to two different vertices

Proposed Communication-Aware DAG model



• Proposed CA-DAG: Communication-Aware DAG model– Two types of vertices: one for computing and one for communications– Edges show define dependences between tasks and order of execution

Jul 1, 2013

32

Communication task

Computing task

• Main advantage– Allows separate resource allocation decisions, assigning

processors to handle computing jobs and network resources for information transmissions

1

4



• Proposed CA-DAG: Communication-Aware DAG model

– Represented by a directed acyclic graph

– Set of vertices is composed of computing tasks and communication tasks

– A computing task is described by a pair with the number of instructions (amount of work) that has to be executed within a specific deadline

– A communication task is described by parameters and defined as the amount of information in bits that has to be successfully transmitted within a predefined deadline

– The set of edges consists of directed edges representing dependence between node and node

Jul 1, 2013



• Example of webmail cloud application– Step 1: Receive user request and process it– Step 2: Generate personalized advertisement– Step 3: Request list of email messages from database– Step 4: Generate HTML pages and send it to the user

Jul 1, 2013


CA-DAG: Communication-Aware DAG

Jul 1, 2013

• Step 1: Receive user request and process it

• Step 2: Generate personalized advertisement

• Step 3: Request list of email messages from database

• Step 4: Generate HTML pages and send it to the user

Step

1

Step

2

Step

3

Step

4

0Receive User

Request

1Process

User Data

2Analyze user

social profile

4Request

Database

3Retrieve

Personalized Ad

5Generate

list of email messages

6Group

conversations

7Generate

HTML page

8Send output

to user

Dzmitry Kliazovich ([email protected])


Jul 1, 201319

Step

1

Step

2

Step

3

Step

4ProcessorNetwork

0Receive User

Request

1Process

User Data

2Analyze user

social profile

4Receive User

Request

3Receive

User Request

5Generate


6Group

conversations

7Generate

HTML page

8Send output

to user

0Receive User

Request

1Process

User Data

2Analyze user

social profile

4Request

Database

3Retrieve

Personalized Ad

5Generate


6Group

conversations

7Generate

HTML page

8Send output

to user

Makespan: 7

20


Jul 1, 2013 Dzmitry Kliazovich ([email protected])





Step

1

Step

2

Step

3

Step

4

6Group

conversations

• Communication unaware model0

Receive User Request

1Process User

Data

2Analyze user social profile

3Retrieve

Personalized Ad

4Request

Database

5Generate list of email messages

7Generate HTML

page

8Send output to

user

21



Step

1

Step

2

Step

3

Step

4

6Group

conversations

• Communication unaware model

0Receive User

Request

1Process User

Data


3Receive User

Request

4Receive User

Request


7Generate HTML

page

8Send output to

user

ProcessorNetwork0

Receive User Request

1Process User

Data


3Retrieve

Personalized Ad

4Request

Database


6Group

conversations

7Generate HTML

page

8Send output to

user

Makespan: 9

0Receive User

Request

4Receive User

Request

3Receive

User Request

8Send

output to user

22







• Edges-based model

Step

1

Step

2

Step

3

Step

4

1Process

User Data

2Analyze user

social profile

5Generate


6Group

conversations

7Generate

HTML page

0Receive

User Request

4Request

Database

3Retrieve

Personalized Ad

8Send output

to user

23



• Edges-based modelSt

ep 1

Step

2

Step

3

Step

4

1Process

User Data

2Analyze user

social profile

5Generate


6Group

conversations

7Generate

HTML page

0Receive

User Request

4Request

Database

3Retrieve

Personalized Ad

8Send output

to user

ProcessorNetwork

0Receive User

Request

0Receive

User Request

1Process

User Data

2Analyze user

social profile

5Generate


6Group

conversations

4Request

Database

3Retrieve

Personalized Ad

4Request

Database

7Generate

HTML page

8Send output

to user

4Request

Database

Makespan: 8



• Comparison of schedulesCA-DAG model Communication-unaware model Edges-based model

Jul 1, 2013



• Comparison of models’ makespan

# ofProcessors

# ofNetwork links

Communication-unaware model

Edges-based model

Proposed CA-DAG model

1 1 9 8 71 2 9 7 72 1 7 8 7

Achieves minimum makespan with the least resources

Jul 1, 2013

Properties of Communication Tasks/Vertices


Properties of Communication Tasks/Vertices

• Task parallelization

• Multipath routing

• Task completion time

• Available bandwidth

Jul 1, 2013


Task Parallelization

• Each communication task/vertex can be divided into different independent communication tasks that can be executed in parallel

• The smallest size of communication task is one bit as all bits in the message are independent

2

1

3 4

1

3 4

2.1 2.2 2.n…

Jul 1, 2013


Multipath Routing

• Most of existing solutions rely on static network topology and fixed pre-allocation which implies circuit switching and pre-defined routing


CoreNetwork

AggregationNetwork

AccessNetwork

S

Links

10 GE 1 GE

Nodes


• In reality, datacenter networks are packet switched with routing decisions taken at every hop

• The availability of multiple paths is essential to benefit from parallelization property of communication tasks

Jul 1, 2013


Task Completion Time

• Execution of communication task involves– Packet transmissions on multiple links– Sequential processing, variable bitrates

• Communication delay components– Processing delay – Queuing delay – Transmission delay – Propagation delay

n1 n2dproc

dtx dqueue dprop

1

( ).N

i i i icomm proc queue tx prop

i

d d d d d

Jul 1, 2013


Available Bandwidth

• Residual Bandwidth– Bandwidth left unoccupied

• Available bandwidth– Bandwidth that a new flow can obtain (residual bandwidth +

portion of the used bandwidth)

• Utilization performance of communication protocols– TCP throughput

( ) ,MSS

B pRTT p

Used bandwidth

Link capacity

Total capacity(includes buffers)

Jul 1, 2013

Performance of CA-DAG Model


Performance of CA-DAG model

• System Architecture– Only one node can communicate at a time

Jul 1, 2013



• Workloads– Winkler graph generator– DAGs with occasional and frequent communications

• Communication-to-Computation Ratio (CCR)

0.1Computationallyintensive DAGs

1Balanced DAGs

2Communicationintensive DAGs

Jul 1, 2013



• Scheduling Algorithm– Offline (deterministic) scheduling– Zero release time of DAGs– Clairvoyant execution and communication time– Adapted list scheduling is employed– A processor allowing minimum execution time is

selected

Jul 1, 2013



• Scheduling Criteria

– Schedule efficiency

– Approximation factor

Jul 1, 2013



• Schedule efficiency

– Ratio of sequential execution time to the makespan of the schedule by the number of computing resources

where is a makespan is the number of computing resources

eff (𝑆 )=∑

𝑖=1. . ,𝑛(𝑝𝑖 )

𝐶𝑚𝑎𝑥×𝑚

Jul 1, 2013


0.1 1 20

10

20

30

40

50

Communication-to-Computation Ratio (CCR)E

ffici

ency

(%

)

CA-DAGComm-unaware DAGEdge-based DAG

0.1 1 20

10

20

30

40

50

Communication-to-Computation Ratio (CCR)

Effi

cien

cy (

%)



• Schedule efficiency Apps. with occasional communications Apps. with frequent communications

The

high

er th

e be

tter

Jul 1, 2013



• Approximation factor

where is a makespan is the optimal makespan is a critical path of the graph

𝐶𝑚𝑎𝑥∗ =𝑚𝑎𝑥 {max (𝑏𝑙𝑒𝑣𝑒𝑙 (𝑡𝑖)) ,

∑𝑖=1. . ,𝑛

(𝑝𝑖 )

𝑚 },𝜌=𝐶𝑚𝑎𝑥 /𝐶𝑚𝑎𝑥

∗

Jul 1, 2013



• Approximation factor

0.1 1 20

0.5

1

1.5

2

2.5

Communications-to-Computation Ratio (CCR)

App

roxi

mat

ion

Fac

tor


0.1 1 20

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Communication-to-Computation Ratio (CCR)A

ppro

xim

atio

n F

acto

r


Apps. with occasional communications Apps. with frequent communications

The

low

er th

e be

tter

Jul 1, 2013


Conclusions

• Cloud applications use communication resources excessively

• New communication-aware model of cloud applications, named CA-DAG, is proposed

• CA-DAG includes separate vertices to represent communication processes to allow making separate resource allocation decisions (computing jobs to processors, communication jobs to the network)

• CA-DAG model enables the design of novel solutions with mixed scheduling policies optimized for cloud computing

Jul 1, 2013

Thank you!

Contact information:

Dzmitry KliazovichUniversity of [email protected]

Date post:	19-Jan-2016
Category:	Documents
Upload:	suzan-bell
View:	217 times
Download:	0 times

CA-DAG: Communication-Aware Directed Acyclic Graphs for Modeling Cloud Computing Applications...

Documents