Abstract:Cloud providers deployed dozens of PoPs and data centers globally to serve billions of geo-distributed users. The traffic management at peering edges has become a key capability of cloud network operators to meet the diverse demands of users. With the rapid growth of cloud applications, users have recently announced new performance requirements, e.g., achieving latency as low as possible instead of maintaining a specified delay. The conventional inter-domain bandwidth allocation approach, which aims to reduce the high operating expenditures of bandwidth usage, fails to meet these new requirements. We further reveal that the flow scheduling among PoPs may fail due to the limited link capacity hidden by the cloud private backbone network controller. Therefore we seek a new traffic management at peering edges. We propose a new controller framework, EdgeCross, that satisfies not only users' emerging demands but maintains low operating costs. The large number of fine-grain application-aware flows and the consideration of backbone links' capacity lead to very high complexity of routing computation and verification for the controller. EdgeCross introduces a two-phase operation that first achieves the low-expense bandwidth allocation according to the standard 95th percentile billing model and then allocates specified flows to peering edges based on users' requirements. EdgeCross further reduces large memory consumption by proposing an effective routing table compression approach. The evaluation based on a production network with 16 PoPs has shown that EdgeCross can successfully process the routes of 1 billion flows in 10 seconds, reduce the average delay for performance-sensitive flows by 2 milliseconds compared to traditional BGP, and is able to save the bandwidth cost by 10-26% compared to the state-of-the-art Cascara.

Where is the Traffic Going? A Comparative Study of Clouds Following Different Designs

ON TRAFFIC MANAGEMENT OF ONLINE CLOUD STORAGE SERVICE

3G Meets the Internet: Understanding the Performance of Hierarchical Routing in 3G Networks.

Towards Cost-Effective Cloud Downloading with Tencent Big Data.

Understanding the Latency to Visit Websites in China: an Infrastructure Perspective

Cloud Service Routing in WDM Networks with Awareness on Delay and Link Failure Probability

Network traffic characteristics of hyperscale data centers in the era of cloud applications

Data Center Network Design for Internet-Related Services and Cloud Computing

Mitigating Routing Inefficiencies to Cloud-Storage Providers: A Case Study.

TailCutter: Wisely Cutting Tail Latency in Cloud CDNs under Cost Constraints

Improving BitTorrent Traffic Performance by Exploiting Geographic Locality

A study of traffic, user behavior and pricing policies in a large campus network

Accessing Cloud with Disaggregated Software-Defined Router.

Demystifying Commercial Content Delivery Networks in China

CDNs Meet CN An Empirical Study of CDN Deployments in China.

A Quantitative and Comparative Study of Network-Level Efficiency for Cloud Storage Services

Adaptive Data Transmission in the Cloud.

EdgeCross: Cloud Scale Traffic Management at Peering Edges

CloudGPS: a Scalable and ISP-friendly Server Selection Scheme in Cloud Computing Environments

Online Traffic Allocation Based on Percentile Charging for Practical CDNs

SIFS : A Scalable and ISP-Friendly Server Selection Scheme in Cloud Services