SlideShare a Scribd company logo
Pradeeban Kathiravelu∗†
Marco Chiesa‡
Pedro Marcos§
Marco Canini¶
Luís Veiga∗
∗
INESC-ID Lisboa / Instituto Superior Técnico, Universidade de Lisboa
†
Université catholique de Louvain ‡
KTH §
UFRGS/FURG ¶
KAUST
IFIP Networking 2018.
Zurich, Switzerland. 15th
May, 2018.
1
Moving Bits with a Fleet of
Shared Virtual Routers
Introduction
2/20
● Increasing demand for bandwidth.
● Decreasing bandwidth prices.
● Pricing Disparity. E.g. IP Transit Price, 2014 (per Mbps)
○ USA: 0.94 $
○ Kazakhstan: 15 $
○ Uzbekistan: 347 $
● What about latency?
○ Online gaming.
○ High-frequency trading.
○ Remote surgery.
Motivation
● Cloud providers have a dedicated connectivity.
○ Well-provisioned and maintained network.
○ Increasing number of regions and points of presence.
● Can a network overlay over cloud instances be used as an
alternative connectivity provider?
○ Cost-effectiveness.
○ High-performance.
○ Optional network services.
3/20
Cloud-Assisted Networks
Virtual/overlay networks over cloud environments
4/20
Our Proposal: NetUber
● A third-party virtual connectivity provider with no fixed
infrastructure.
○ An overlay network, leveraging multi-cloud infrastructures.
5/20
NetUber Application Scenarios
1. Cheaper transfers between two endpoints.
2. Higher throughput or reduced latency.
3. Better alternative to SaaS replication.
4. Network services (compression, encryption, ..).
6/20
● Feasibility Study: Platform Cost of NetUber
7/20
A. Cost of Cloud Instances.
○ Charged per second.
○ Very high.
B. Cost of Bandwidth.
○ Charged per data transferred.
○ Also very high.
C. Cost to connect to
the cloud provider.
Scenario (1 of 4): Cheaper Transfers
A) Cost of Cloud Instances: Observations
● 10 Gbps R4 instance (r4.8xlarge) pairs offered only
maximum of 1.2 Gbps of data transfer inter-region.
○ 10 Gbps only inside a placement group.
● We need more pairs
of instances!
8/20
Scenario (1 of 4): Cheaper Transfers
Spot Instances!
● Cheaper (up to 90% savings), but volatile, instances.
● Price Fluctuations - Future price unpredictable (for EC2).
● Differing prices among availability zones of a region.
○ Buy from the cheapest availability zones at the moment.
○ Maintain instances in the cheap availability zones.
9/22
Scenario (1 of 4): Cheaper Transfers
B) Cost of Bandwidth: Price disparity is real!
10/20
● Regions 1 - 9 (US, Canada, and EU) remain much cheaper
than the others.
Scenario (1 of 4): Cheaper Transfers
C) Cost to connect to the cloud provider
11/20
● Connect the end-user to the cloud servers.
● Often provided by the cloud provider.
○ Example: Amazon Direct Connect.
○ Charged per port-hour (e.g. how many hours a 10 GbE port is used).
Scenario (1 of 4): Cheaper Transfers
Cloud-Assisted Point-to-Point Connectivity
12/20
● Also cheaper than MPLS networks or transit providers.
○ Thanks to spot instances.
Scenario (2 of 4): Higher throughput or reduced latency
● Better control over the path, compared to the Internet paths.
13/20
Scenario (3 of 4): Better Alternative to SaaS Replication
● Deploy Software-as-a-Service (SaaS) applications in just one region.
○ Use NetUber to access them from another region.
■ Instead of replicating them across multiple cloud regions.
● Access to more regions by leveraging multiple cloud providers.
14/20
Scenario (4 of 4): Network Services
● NetUber uses memory-optimized R4 spot instances.
○ Each instance with 244 GB memory, 32 vCPU, and 10 GbE interface.
● Possibility to deploy network services at the instances.
● Network services.
○ Value-added services for the customer.
■ Encryption, WAN-Optimizer, load balancer, ..
○ Services for cost-efficiency.
■ Compression.
Evaluation
● Cheaper point-to-point connectivity.
○ AWS as the overlay cloud provider.
○ Compared against a transit provider and another connectivity provider
with a large global backbone network.
● Improve latency with cloud routes.
○ Compared to ISPs.
○ Traffic sent from: RIPE Atlas Probes and distributed servers.
○ Destination: AWS distributed servers from the AWS regions.
○ ISPs vs. ISP to the nearest AWS region and then NetUber overlay.
15/20
1) Cheaper point-to-point connectivity
16/20
● Expense for 10 Gbps flat connectivity
○ Measured for transfers from EU and USA.
○ Cheaper for data transfers <50 TB.
2) Improve latency with cloud routes
17/20
● Instead of sending traffic A -> Z, can we send A -> B -> Z?
○ B is closer to A. B and Z are servers in cloud regions.
○ B and Z are connected by NetUber overlay.
Ping times: ISP vs. NetUber (via region, % improvement)
18/20
● NetUber cuts Internet latencies up to a factor of 30%.
● The use of Direct Connect would make this even better.
Related Work
● Industrial efforts on infrastructure to offer connectivity.
○ Teridion - Internet fast lanes for SaaS providers.
○ Voxility - Large scale globally distributed infrastructure as an alternative
to transit providers.
● Previous research focus on technical side.
○ Not economical aspects - More expensive.
○ NetUber as a cheaper alternative, with spot instances.
19/20
Conclusion
● A connectivity provider that does not own the infrastructure.
● “Internet Fast-routes” through cloud-assisted networks.
○ Better than ISPs (~50 - 75 Mbps, often with a cap) for end-users.
● Cheaper point-to-point connectivity.
○ Cheaper than transit providers and similar offerings (for < 50 TB/month).
● Future work:
○ Evaluate NetUber for more parameters (loss rate, jitter, ..)
○ Evaluate the cost with more cloud providers and pairs of regions.
20/20
Conclusion
21/21
Thank you!
● A connectivity provider that does not own the infrastructure.
● “Internet Fast-routes” through cloud-assisted networks.
○ Better than ISPs (~50 - 75 Mbps, often with a cap) for end-users.
● Cheaper point-to-point connectivity.
○ Cheaper than transit providers and similar offerings (for < 50 TB/month).
● Future work:
○ Evaluate NetUber for more parameters (loss rate, jitter, ..)
○ Evaluate the cost with more cloud providers and pairs of regions.

More Related Content

PDF
Software-Defined Systems for Network-Aware Service Composition and Workflow P...
PDF
Software-Defined Data Services: Interoperable and Network-Aware Big Data Exec...
PPTX
PacketCloud: an Open Platform for Elastic In-network Services.
PDF
Software-Defined Inter-Cloud Composition of Big Services
PDF
6. The grid-COMPUTING OGSA and WSRF
PDF
Hitachi datasheet-universal-replicator
PDF
A location based least-cost scheduling for data-intensive applications
PDF
F233842
Software-Defined Systems for Network-Aware Service Composition and Workflow P...
Software-Defined Data Services: Interoperable and Network-Aware Big Data Exec...
PacketCloud: an Open Platform for Elastic In-network Services.
Software-Defined Inter-Cloud Composition of Big Services
6. The grid-COMPUTING OGSA and WSRF
Hitachi datasheet-universal-replicator
A location based least-cost scheduling for data-intensive applications
F233842

What's hot (18)

PPTX
An assessment of internet of things protocols for constrain apps
PPT
Lambda Data Grid
PDF
Content centric networks
PPTX
Overlay networks ppt
PPTX
Named data networking
PDF
Ieeepro techno solutions 2014 ieee java project - cloud bandwidth and cost ...
PDF
Dynamic adaptation balman
PPTX
Job sequence scheduling for cloud computing
PDF
Named data networking. Basic Principle
PDF
Route Server Peering Improves End User "Quality of Experience"
PPTX
QoS-Aware Data Replication for Data-Intensive Applications in Cloud Computing...
DOCX
ORCHESTRATING BULK DATA TRANSFERS ACROSS GEO-DISTRIBUTED DATACENTERS
PPTX
Faster Content Distribution with Content Addressable NDN Repository
PDF
WRNP18 - Software Defined Infrastructures: Multi-Domain Orchestration
PDF
DOC
Distributed, concurrent, and independent access to encrypted cloud databases
PPTX
mqtt intro short
PPT
Distributed, concurrent, and independent access to encrypted cloud databases
An assessment of internet of things protocols for constrain apps
Lambda Data Grid
Content centric networks
Overlay networks ppt
Named data networking
Ieeepro techno solutions 2014 ieee java project - cloud bandwidth and cost ...
Dynamic adaptation balman
Job sequence scheduling for cloud computing
Named data networking. Basic Principle
Route Server Peering Improves End User "Quality of Experience"
QoS-Aware Data Replication for Data-Intensive Applications in Cloud Computing...
ORCHESTRATING BULK DATA TRANSFERS ACROSS GEO-DISTRIBUTED DATACENTERS
Faster Content Distribution with Content Addressable NDN Repository
WRNP18 - Software Defined Infrastructures: Multi-Domain Orchestration
Distributed, concurrent, and independent access to encrypted cloud databases
mqtt intro short
Distributed, concurrent, and independent access to encrypted cloud databases
Ad

Similar to Moving bits with a fleet of shared virtual routers (20)

PDF
The UCLouvain Public Defense of my EMJD-DC Double Doctorate Ph.D. degree
PDF
My Ph.D. Defense - Software-Defined Systems for Network-Aware Service Composi...
PDF
Internet Exchange Points, presented at Peering Workshop at the PITA 29th AGM,...
PPTX
Multi-Cluster Load Balancing in Kubernetes_ Strategies and Considerations.pptx
PDF
PacNOG 31: Internet Exchange Points
PDF
PITA 27th AGM & Business Forum Expo 23: Internet Exchange Points
PPTX
Cloud interconnection networks basic .pptx
PDF
Big Data Transport
PDF
UCL Ph.D. Confirmation 2018
PPTX
QoS.pptx
PDF
Networking in Cloud Computing Environment
PDF
WINS: Peering and IXPs
PDF
Experimental Evaluation of Large Scale WiFi Multicast Rate Control, By: Varun...
PPTX
Keeping the Internet Fast and Resilient for You and Your Customers
PPTX
Broad Sky SD-WAN September 2018
PDF
Bench, a Framework for Benchmarking Kafka Using K8s and OpenMessaging Benchma...
PPTX
Lecture notes - Data Centers________.pptx
PDF
Better Than Best Effort at Bloomberg from ThousandEyes Connect
PDF
ISP Network Design workshops how to design networks
PDF
5 maximazing networkcapacity_v4-jorge_alvarado
The UCLouvain Public Defense of my EMJD-DC Double Doctorate Ph.D. degree
My Ph.D. Defense - Software-Defined Systems for Network-Aware Service Composi...
Internet Exchange Points, presented at Peering Workshop at the PITA 29th AGM,...
Multi-Cluster Load Balancing in Kubernetes_ Strategies and Considerations.pptx
PacNOG 31: Internet Exchange Points
PITA 27th AGM & Business Forum Expo 23: Internet Exchange Points
Cloud interconnection networks basic .pptx
Big Data Transport
UCL Ph.D. Confirmation 2018
QoS.pptx
Networking in Cloud Computing Environment
WINS: Peering and IXPs
Experimental Evaluation of Large Scale WiFi Multicast Rate Control, By: Varun...
Keeping the Internet Fast and Resilient for You and Your Customers
Broad Sky SD-WAN September 2018
Bench, a Framework for Benchmarking Kafka Using K8s and OpenMessaging Benchma...
Lecture notes - Data Centers________.pptx
Better Than Best Effort at Bloomberg from ThousandEyes Connect
ISP Network Design workshops how to design networks
5 maximazing networkcapacity_v4-jorge_alvarado
Ad

More from Pradeeban Kathiravelu, Ph.D. (20)

PDF
Google Summer of Code_2023.pdf
PDF
Google Summer of Code (GSoC) 2022
PDF
Google Summer of Code (GSoC) 2022
PPTX
Niffler: A DICOM Framework for Machine Learning and Processing Pipelines.
PDF
Google summer of code (GSoC) 2021
PPTX
A DICOM Framework for Machine Learning Pipelines against Real-Time Radiology ...
PDF
Google Summer of Code (GSoC) 2020 for mentors
PDF
Google Summer of Code (GSoC) 2020
PDF
Data Services with Bindaas: RESTful Interfaces for Diverse Data Sources
PDF
My Ph.D. Defense - Software-Defined Systems for Network-Aware Service Compos...
PDF
On-Demand Service-Based Big Data Integration: Optimized for Research Collabor...
PDF
Scalability and Resilience of Multi-Tenant Distributed Clouds in the Big Serv...
PDF
Scalability and Resilience of Multi-Tenant Distributed Clouds in the Big Serv...
PDF
Componentizing Big Services in the Internet
PDF
SD-CPS: Taming the Challenges of Cyber-Physical Systems with a Software-Defin...
PDF
ViTeNA: An SDN-Based Virtual Network Embedding Algorithm for Multi-Tenant Dat...
PDF
Software-Defined Simulations for Continuous Development of Cloud and Data Cen...
PDF
Selective Redundancy in Network-as-a-Service: Differentiated QoS in Multi-Ten...
PDF
Building Blocks of Mayan: Componentizing the eScience Workflows Through Softw...
PDF
Software-Defined Approach for QoS and Data Quality in Multi-Tenant Clouds
Google Summer of Code_2023.pdf
Google Summer of Code (GSoC) 2022
Google Summer of Code (GSoC) 2022
Niffler: A DICOM Framework for Machine Learning and Processing Pipelines.
Google summer of code (GSoC) 2021
A DICOM Framework for Machine Learning Pipelines against Real-Time Radiology ...
Google Summer of Code (GSoC) 2020 for mentors
Google Summer of Code (GSoC) 2020
Data Services with Bindaas: RESTful Interfaces for Diverse Data Sources
My Ph.D. Defense - Software-Defined Systems for Network-Aware Service Compos...
On-Demand Service-Based Big Data Integration: Optimized for Research Collabor...
Scalability and Resilience of Multi-Tenant Distributed Clouds in the Big Serv...
Scalability and Resilience of Multi-Tenant Distributed Clouds in the Big Serv...
Componentizing Big Services in the Internet
SD-CPS: Taming the Challenges of Cyber-Physical Systems with a Software-Defin...
ViTeNA: An SDN-Based Virtual Network Embedding Algorithm for Multi-Tenant Dat...
Software-Defined Simulations for Continuous Development of Cloud and Data Cen...
Selective Redundancy in Network-as-a-Service: Differentiated QoS in Multi-Ten...
Building Blocks of Mayan: Componentizing the eScience Workflows Through Softw...
Software-Defined Approach for QoS and Data Quality in Multi-Tenant Clouds

Recently uploaded (20)

PDF
RMMM.pdf make it easy to upload and study
PDF
Chapter 2 Heredity, Prenatal Development, and Birth.pdf
PPTX
master seminar digital applications in india
PPTX
1st Inaugural Professorial Lecture held on 19th February 2020 (Governance and...
PPTX
school management -TNTEU- B.Ed., Semester II Unit 1.pptx
PPTX
Microbial diseases, their pathogenesis and prophylaxis
PDF
01-Introduction-to-Information-Management.pdf
PDF
2.FourierTransform-ShortQuestionswithAnswers.pdf
PPTX
Final Presentation General Medicine 03-08-2024.pptx
PPTX
GDM (1) (1).pptx small presentation for students
PPTX
Cell Structure & Organelles in detailed.
PDF
102 student loan defaulters named and shamed – Is someone you know on the list?
PDF
Classroom Observation Tools for Teachers
PDF
grade 11-chemistry_fetena_net_5883.pdf teacher guide for all student
PDF
ANTIBIOTICS.pptx.pdf………………… xxxxxxxxxxxxx
PDF
GENETICS IN BIOLOGY IN SECONDARY LEVEL FORM 3
PDF
The Lost Whites of Pakistan by Jahanzaib Mughal.pdf
PPTX
Presentation on HIE in infants and its manifestations
PDF
3rd Neelam Sanjeevareddy Memorial Lecture.pdf
PPTX
Final Presentation General Medicine 03-08-2024.pptx
RMMM.pdf make it easy to upload and study
Chapter 2 Heredity, Prenatal Development, and Birth.pdf
master seminar digital applications in india
1st Inaugural Professorial Lecture held on 19th February 2020 (Governance and...
school management -TNTEU- B.Ed., Semester II Unit 1.pptx
Microbial diseases, their pathogenesis and prophylaxis
01-Introduction-to-Information-Management.pdf
2.FourierTransform-ShortQuestionswithAnswers.pdf
Final Presentation General Medicine 03-08-2024.pptx
GDM (1) (1).pptx small presentation for students
Cell Structure & Organelles in detailed.
102 student loan defaulters named and shamed – Is someone you know on the list?
Classroom Observation Tools for Teachers
grade 11-chemistry_fetena_net_5883.pdf teacher guide for all student
ANTIBIOTICS.pptx.pdf………………… xxxxxxxxxxxxx
GENETICS IN BIOLOGY IN SECONDARY LEVEL FORM 3
The Lost Whites of Pakistan by Jahanzaib Mughal.pdf
Presentation on HIE in infants and its manifestations
3rd Neelam Sanjeevareddy Memorial Lecture.pdf
Final Presentation General Medicine 03-08-2024.pptx

Moving bits with a fleet of shared virtual routers

  • 1. Pradeeban Kathiravelu∗† Marco Chiesa‡ Pedro Marcos§ Marco Canini¶ Luís Veiga∗ ∗ INESC-ID Lisboa / Instituto Superior Técnico, Universidade de Lisboa † Université catholique de Louvain ‡ KTH § UFRGS/FURG ¶ KAUST IFIP Networking 2018. Zurich, Switzerland. 15th May, 2018. 1 Moving Bits with a Fleet of Shared Virtual Routers
  • 2. Introduction 2/20 ● Increasing demand for bandwidth. ● Decreasing bandwidth prices. ● Pricing Disparity. E.g. IP Transit Price, 2014 (per Mbps) ○ USA: 0.94 $ ○ Kazakhstan: 15 $ ○ Uzbekistan: 347 $ ● What about latency? ○ Online gaming. ○ High-frequency trading. ○ Remote surgery.
  • 3. Motivation ● Cloud providers have a dedicated connectivity. ○ Well-provisioned and maintained network. ○ Increasing number of regions and points of presence. ● Can a network overlay over cloud instances be used as an alternative connectivity provider? ○ Cost-effectiveness. ○ High-performance. ○ Optional network services. 3/20
  • 5. Our Proposal: NetUber ● A third-party virtual connectivity provider with no fixed infrastructure. ○ An overlay network, leveraging multi-cloud infrastructures. 5/20
  • 6. NetUber Application Scenarios 1. Cheaper transfers between two endpoints. 2. Higher throughput or reduced latency. 3. Better alternative to SaaS replication. 4. Network services (compression, encryption, ..). 6/20
  • 7. ● Feasibility Study: Platform Cost of NetUber 7/20 A. Cost of Cloud Instances. ○ Charged per second. ○ Very high. B. Cost of Bandwidth. ○ Charged per data transferred. ○ Also very high. C. Cost to connect to the cloud provider. Scenario (1 of 4): Cheaper Transfers
  • 8. A) Cost of Cloud Instances: Observations ● 10 Gbps R4 instance (r4.8xlarge) pairs offered only maximum of 1.2 Gbps of data transfer inter-region. ○ 10 Gbps only inside a placement group. ● We need more pairs of instances! 8/20 Scenario (1 of 4): Cheaper Transfers
  • 9. Spot Instances! ● Cheaper (up to 90% savings), but volatile, instances. ● Price Fluctuations - Future price unpredictable (for EC2). ● Differing prices among availability zones of a region. ○ Buy from the cheapest availability zones at the moment. ○ Maintain instances in the cheap availability zones. 9/22 Scenario (1 of 4): Cheaper Transfers
  • 10. B) Cost of Bandwidth: Price disparity is real! 10/20 ● Regions 1 - 9 (US, Canada, and EU) remain much cheaper than the others. Scenario (1 of 4): Cheaper Transfers
  • 11. C) Cost to connect to the cloud provider 11/20 ● Connect the end-user to the cloud servers. ● Often provided by the cloud provider. ○ Example: Amazon Direct Connect. ○ Charged per port-hour (e.g. how many hours a 10 GbE port is used). Scenario (1 of 4): Cheaper Transfers
  • 12. Cloud-Assisted Point-to-Point Connectivity 12/20 ● Also cheaper than MPLS networks or transit providers. ○ Thanks to spot instances. Scenario (2 of 4): Higher throughput or reduced latency ● Better control over the path, compared to the Internet paths.
  • 13. 13/20 Scenario (3 of 4): Better Alternative to SaaS Replication ● Deploy Software-as-a-Service (SaaS) applications in just one region. ○ Use NetUber to access them from another region. ■ Instead of replicating them across multiple cloud regions. ● Access to more regions by leveraging multiple cloud providers.
  • 14. 14/20 Scenario (4 of 4): Network Services ● NetUber uses memory-optimized R4 spot instances. ○ Each instance with 244 GB memory, 32 vCPU, and 10 GbE interface. ● Possibility to deploy network services at the instances. ● Network services. ○ Value-added services for the customer. ■ Encryption, WAN-Optimizer, load balancer, .. ○ Services for cost-efficiency. ■ Compression.
  • 15. Evaluation ● Cheaper point-to-point connectivity. ○ AWS as the overlay cloud provider. ○ Compared against a transit provider and another connectivity provider with a large global backbone network. ● Improve latency with cloud routes. ○ Compared to ISPs. ○ Traffic sent from: RIPE Atlas Probes and distributed servers. ○ Destination: AWS distributed servers from the AWS regions. ○ ISPs vs. ISP to the nearest AWS region and then NetUber overlay. 15/20
  • 16. 1) Cheaper point-to-point connectivity 16/20 ● Expense for 10 Gbps flat connectivity ○ Measured for transfers from EU and USA. ○ Cheaper for data transfers <50 TB.
  • 17. 2) Improve latency with cloud routes 17/20 ● Instead of sending traffic A -> Z, can we send A -> B -> Z? ○ B is closer to A. B and Z are servers in cloud regions. ○ B and Z are connected by NetUber overlay.
  • 18. Ping times: ISP vs. NetUber (via region, % improvement) 18/20 ● NetUber cuts Internet latencies up to a factor of 30%. ● The use of Direct Connect would make this even better.
  • 19. Related Work ● Industrial efforts on infrastructure to offer connectivity. ○ Teridion - Internet fast lanes for SaaS providers. ○ Voxility - Large scale globally distributed infrastructure as an alternative to transit providers. ● Previous research focus on technical side. ○ Not economical aspects - More expensive. ○ NetUber as a cheaper alternative, with spot instances. 19/20
  • 20. Conclusion ● A connectivity provider that does not own the infrastructure. ● “Internet Fast-routes” through cloud-assisted networks. ○ Better than ISPs (~50 - 75 Mbps, often with a cap) for end-users. ● Cheaper point-to-point connectivity. ○ Cheaper than transit providers and similar offerings (for < 50 TB/month). ● Future work: ○ Evaluate NetUber for more parameters (loss rate, jitter, ..) ○ Evaluate the cost with more cloud providers and pairs of regions. 20/20
  • 21. Conclusion 21/21 Thank you! ● A connectivity provider that does not own the infrastructure. ● “Internet Fast-routes” through cloud-assisted networks. ○ Better than ISPs (~50 - 75 Mbps, often with a cap) for end-users. ● Cheaper point-to-point connectivity. ○ Cheaper than transit providers and similar offerings (for < 50 TB/month). ● Future work: ○ Evaluate NetUber for more parameters (loss rate, jitter, ..) ○ Evaluate the cost with more cloud providers and pairs of regions.