network systems laboratory research today there is no ...joe tucci , emc ceo ^s] µooÇ À Ç v ]...

1
Network Systems Laboratory Research Xin Sun Assistant Professor, SCIS On-going Project 1: Network Complexity Motivation: Human errors are the number 1 cause of network problems Operators ability to manage a network decreases as the network become more complex Design faults and configurations are the number 1 cause of network outages and cypher-attacks. Today there is no scientific way to reason about and control network complexity. Vision: complexity must be an explicit objective when designing/evolving a network For the same target network, many designs are possible But different designs will result in significantly different complexity Need scientific models to guide operators to identify the optimal design. Modeling Details (Illustrated with Route Redistribution) EIGRP 10 OSPF 20 1. router ospf 20 2. redistribute eigrp 10 3. ! 4. router eigrp 10 5. redistribute ospf 20 route-map EIGRP2OSPF 6. ! 7. route-map EIGRP2OSPF permit 10 8. match ip address 1 9. ! 10. access-list 1 permit s4 11. ! Total complexity (in one direction) = complexity of configuring the RR itself + Complexity of configuring the route filter (if needed) OSPF EIGRP OSPF - all EIGRP s4 - (i,j): the subnets in instance j that instance i can reach S3 S1 S2 S4 Instance-level reachability matrix More info in the paper: “Modeling Complexity of Enterprise Routing design”, Xin Sun, Sanjay Rao and Geoffrey Xie, ACM CoNEXT, 2012. And there are some things they might not want to put in the cloud for security and reliability reasons….So, you've got to have these kinds of hybrid solutions.” Steve Ballmer, Microsoft CEO “We think it's a combination of putting applications in your own data center, and then use the cloud to take out peaks, or you could put specific things in the cloud.” Joe Tucci, EMC CEO “Virtually every enterprise will adopt a hybrid format” Russ Daniels, CTO of cloud computing, HP On-going Project 2: Enabling Wide Adoption of Cloud Computing Typical Enterprise Application Architecture Evaluation Results Using a Real Enterprise ERP Application 15% w/ policy $14K FE1(1) BL1(2),BL2,BL4,BL5 Conclusion: hybrid clouds can achieve cost savings while meeting enterprise policies and delay bounds Recommended components to migrate Bound on increase in mean delay Yearly Savings FE BL BE 30% $58K migrate all components in full 15% $38K FE1(1),FE2 BL1(1),BL2,BL3,BL4,BL5 BE2,BE3,BE4,BE5 More info in the paper: “Cloudward Bound: Planning for Beneficial Migration of Enterprise Applications to the Cloud”, Mohammad Hajjat, Xin Sun, et al., ACM SIGCOMM, 2010. Research Deliverables: Software Tools Released Toolkit #1 for visualizing enterprise routing design Toolkit #2 for automating & visualizing VLAN management tasks Our Approach: A Top-Down Analytic Framework for Modeling Complexity network design device configurations Implemented by realized by manager intent Decomposes into individual components Ensures the design meets the intent Sheds light on how each component contribute to complexity Estimates # of referential links in the resulting configurations Models intent Enables “what- if” analysis Ensures correctness of the design. Research Overview Tackling emerging problems in enterprise networks, cloud computing, and software defined networking. Emphasize on network design, management and migration. “From practice, to theory, back to practice". Identifying problems through interactions with the industry and network operators community. Solving problems by developing scientifically-sound models and algorithms Transferring solutions to deployable software tools

Upload: others

Post on 22-Sep-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Network Systems Laboratory Research Today there is no ...Joe Tucci , EMC CEO ^s] µooÇ À Ç v ] Á]oo } ZÇ ] ( } u _ Russ Daniels, CTO of cloud computing, HP On -going Project 2:

Network Systems Laboratory Research

Xin Sun

Assistant Professor, SCIS

On-going Project 1: Network Complexity

• Motivation: Human errors are the number 1 cause of network problems– Operators ability to manage a network decreases as the network

become more complex– Design faults and configurations are the number 1 cause of network

outages and cypher-attacks.– Today there is no scientific way to reason about and control network

complexity.

• Vision: complexity must be an explicit objective when designing/evolving a network– For the same target network, many designs are possible– But different designs will result in significantly different complexity– Need scientific models to guide operators to identify the optimal

design.

Modeling Details (Illustrated withRoute Redistribution)

EIGRP 10 OSPF 20

1. router ospf 202. redistribute eigrp 103. !4. router eigrp 105. redistribute ospf 20 route-map EIGRP2OSPF6. !7. route-map EIGRP2OSPF permit 108. match ip address 19. !10. access-list 1 permit s411. !

• Total complexity (in one direction) =• complexity of configuring

the RR itself +• Complexity of configuring

the route filter (if needed)

OSPF EIGRP

OSPF - all

EIGRP s4 -

(i,j): the subnets in instance j that instance i can reach

S3S1

S2

S4

Instance-level reachability matrix

More info in the paper: “Modeling Complexity of Enterprise Routing design”, Xin Sun, Sanjay Rao and Geoffrey Xie, ACM CoNEXT, 2012.

“And there are some things they might not want to put in the cloud for security and reliability reasons….So, you've got to have these kinds of hybrid solutions.”

Steve Ballmer, Microsoft CEO

“We think it's a combination of putting applications in your own data center, and then use the cloud to take out peaks, or you could put specific things in the cloud.”

Joe Tucci, EMC CEO

“Virtually every enterprise will adopt a hybrid format”

Russ Daniels, CTO of cloud computing, HP

On-going Project 2: Enabling Wide Adoption of Cloud Computing

Typical Enterprise Application Architecture

Evaluation Results Using a Real Enterprise ERP Application

15% w/ policy $14K FE1(1) BL1(2),BL2,BL4,BL5 …

• Conclusion: hybrid clouds can achieve cost savings while meeting enterprise policies and delay bounds

Recommended components to migrate

Bound on increase in mean delay

Yearly Savings

FE BL BE

30% $58K migrate all components in full

15% $38K FE1(1),FE2 BL1(1),BL2,BL3,BL4,BL5 BE2,BE3,BE4,BE5

More info in the paper: “Cloudward Bound: Planning for Beneficial Migration of Enterprise Applications to the Cloud”, Mohammad Hajjat, Xin Sun, et al., ACM SIGCOMM, 2010.

Research Deliverables: Software Tools ReleasedToolkit #1 for visualizing enterprise routing design

Toolkit #2 for automating & visualizing VLAN management tasks

Our Approach: A Top-Down Analytic Framework for Modeling Complexity

network design

device configurations

Implemented by

realized by

manager intent

Decomposes into individual components

Ensures the design meets the intent

Sheds light on how each component contribute to complexity

Estimates # of referential links in the resulting configurations

Models intent

Enables “what-if” analysis

Ensures correctness of the design.

Research Overview

• Tackling emerging problems in enterprise networks, cloud computing, and software defined networking.– Emphasize on network design, management and

migration.

• “From practice, to theory, back to practice". – Identifying problems through interactions with the

industry and network operators community.

– Solving problems by developing scientifically-sound models and algorithms

– Transferring solutions to deployable software tools