Loops More loops Off-by-one errors Infinite loops Nested Loops.
Network utilization with SDN in on-demand application ... loops that observe and modify the behavior...
-
Upload
nguyenmien -
Category
Documents
-
view
223 -
download
3
Transcript of Network utilization with SDN in on-demand application ... loops that observe and modify the behavior...
Introduction Theoretical Framework Implementation Findings and Conclusions
Network utilization with SDN in on-demandapplication-specific networks
Ioannis [email protected]
Supervised by:Marc X. Makkes
System and Network EngineeringUniversiteit van Amsterdam
July 1, 2015
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 1 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Internet factories
Internet factories: Creating application-specific networks on-demand[1]Uses Infastructure-as-a-Service cloudsCreate, configure and modify the infastructureSecond implementation Compute factory
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 2 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Overlay networks
A network built on top of one or more existing networks
Add extra functionality
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 3 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
OSPF / SDN comparison
Open Shortest Path First (OSPF) :
Mature protocol
Widely used and supported
Uses Dijkstra’s algorithm
Used by Compute factory
Software Defined Networking (SDN):
Separation between control plane anddata plane
Centralized managment
Programmability
Routing granularity
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 4 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Hypothesis
If the created overlay networks make use of SDN (OpenFlow), Compute factory’scontrol loops that observe and modify the behavior can gain benefits.
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 5 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Related work
B4: Experience with a Globally-Deployed Software Defined WAN[2]
MiceTrap: Scalable Traffic Engineering of Datacenter Mice Flows usingOpenFlow[3]
SDN Based Load Balancing Mechanism for Elephant Flow in Data CenterNetworks[4]
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 6 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Differences from our case
Virtual Machine migration
Connection speed
Dynamic infrastructure
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 7 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Elephants and Mice flows
Elephant flow: Long-lived flowwith large data transfer
Mice flow: Short-lived flowwith small data transfer
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 8 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Compute factory
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 9 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Compute factory flow control loop
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 10 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Scenarios
FirstTransfer sequential small and large file in empty path
SecondTransfer simultaneously small and large file with the Copmute factorycontrol loop disabled
ThirdTransfer simultaneously small and large file with the Copmute factorycontrol loop enabled
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 11 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Scenario results
Total time transferring a file
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 12 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
CPU utilization
CPU utilization in the intermediate switches
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 13 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Conclusions
Increase stability in data transfer
Dicrease jitter
Balance the CPU load in intermediate switches
Not increase network utilization
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 14 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
Thank you
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 15 / 16
Introduction Theoretical Framework Implementation Findings and Conclusions
References I
Rudolf Strijkers, Marc X. Makkes, Cees de Laat, Robert MeijerInternet factories: Creating application-specific networks on-demandComputer Networks, 68:187-198, 2014.
Sushant Jain et al.B4: Experience with a Globally-Deployed Software Defined WANACM SIGCOMM, 3-14, 2013.
Trestian R., Muntean G.-M., Katrinis K.MiceTrap: Scalable Traffic Engineering of Datacenter Mice Flows using OpenFlowIntegrated Network Management, 904-907 , 2013.
Jing Liu, Jie Li, Guochu Shou, Yihong Hu, Zhigang Guo, Wei DaiSDN Based Load Balancing Mechanism for Elephant Flow in Data CenterNetworksWireless Personal Multimedia Communications, 486-490, 2014.
Ioannis Grafis Universiteit van Amsterdam July 1, 2015 16 / 16