Live migration, the process of transferring a running application to a different physical location with minimal downtime, can provide many benefits desired by modern cloudbased systems. Furthermore, live migration between different cloud providers enables a new level of freedom for cloud users to move their workloads around for performance or business objectives without having to be tied down to any single provider. While this vision is not new, to-date, there are few solutions and proof-of-concepts that provide this capability. As containerized applications are gaining popularity, we focus on the design and implementation of live migration of containers across cloud providers. CloudHopper, our proof-of-concept live migration service for containers to hop around between Amazon Web Services, Google Cloud Platform, and Microsoft Azure is evaluated using a common web-based workload. CloudHopper is automated and supports pre-copy optimization, connection holding, traffic redirection, and multiple interdependent container migration. It is applicable to a broad range of application use cases.

Multi-tier applications are widely deployed in today's virtualized cloud computing environments. At the same time, management operations in these virtualized environments, such as load balancing, hardware maintenance, workload consolidation, etc., often make use of live virtual machine (VM) migration to control the placement of VMs. Although existing solutions are able to migrate a single VM efficiently, little attention has been devoted to migrating related VMs in multi-tier applications. Ignoring the relatedness of VMs during migration can lead to serious application performance degradation. This paper formulates the multi-tier application migration problem, and presents a new communication-impact-driven coordinated approach, as well as a system called COMMA that realizes this approach. Through extensive testbed experiments, numerical analyses, and a demonstration of COMMA on Amazon EC2, we show that this approach is highly effective in minimizing migration's impact on multi-tier applications' performance.

Deploying interactive applications in the cloud is a challenge due to the high variability in performance of cloud services. In this paper, we present Dealer — a system that helps geo-distributed, interactive and multi-tier applications meet their stringent requirements on response time despite such variability. Our approach is motivated by the fact that, at any time, only a small number of application components of large multi-tier applications experience poor performance. Dealer continually monitors the performance of individual components and communication latencies between them to build a global view of the application. In serving any given request, Dealer seeks to minimize user response times by picking the best combination of replicas (potentially located across different data centers). While Dealer requires modifications to application code, we show the changes required are modest. Our evaluations on two multi-tier applications using real cloud deployments indicate the 90%ile of response times could be reduced by more than a factor of 6 under natural cloud dynamics. Our results indicate the cost of inter-data-center traffic with Dealer is minor, and that Dealer can in fact be used to reduce the overall operational costs of applications by up to 15% by leveraging the difference in billing plans of cloud instances.

Live migration of virtual machines is a key management function in cloud computing. Unfortunately, no live migration progress management system exists in the state-ofthe- art, leading to (1) guesswork over how long a migration might take and the inability to schedule dependent tasks accordingly; (2) unacceptable application degradation when application components become split over distant cloud datacenters for an arbitrary period during migration; (3) inability to tradeoff application performance and migration time e.g. to finish migration later for less impact on application performance. Pacer is the first migration progress management system that solves these problems. Pacer's techniques are based on robust and lightweight run-time measurements of system and workload characteristics, efficient and accurate analytic models for progress predictions, and online adaptation to maintain user-defined migration objectives for coordinated and timely migrations. Our experiments on a local testbed and on Amazon EC2 show that Pacer is highly effective under a range of application workloads and network conditions.

Deploying interactive applications in the cloud is a challenge due to the high variability in performance of cloud services. In this paper, we present Dealer-- a system that helps geo-distributed, interactive and multi-tier applications meet their stringent requirements on response time despite such variability. Our approach is motivated by the fact that, at any time, only a small number of application components of large multi-tier applications experience poor performance. Dealer abstracts application structure as a component graph, with nodes being application components and edges capturing inter-component communication patterns. Dealer continually monitors the performance of individual component replicas and communication latencies between replica pairs. In serving any given user request, Dealer seeks to minimize user response times by picking the best combination of replicas (potentially located across different data-centers). While Dealer does require modifications to application code, we show through integration with two multi-tier applications that the changes required are modest. Our evaluations on two multi-tier applications using real cloud deployments indicate the 90%ile of application response times could be reduced by a factor of 3 under natural cloud dynamics compared to conventional data-center redirection techniques which are agnostic of application structure.

Cloud computing carries the promise of powerful models and abstractions that could transform the way IT services are delivered today. While the cloud's economic model has been argued to provide cost benefits compared to traditional in-house IT delivery, it remains to be seen how this argument changes for different business applications and environments. In this paper, we develop a framework for exploring the factors that influence IT delivery cost within an organization compared to the cost of using the cloud. We then apply the framework to develop insight on the economics of migrating to the cloud from the perspective of users in emerging markets. We find that emerging markets have different characteristics from established economies that result in different barriers and benefits realized from using the cloud.

The emerging open cloud computing model will provide users with great freedom to dynamically migrate virtualized computing services to, from, and between clouds over the wide-area. While this freedom leads to many potential benefits, the running services must be minimally disrupted by the migration. Unfortunately, current solutions for wide-area migration incur too much disruption as they will significantly slow down storage I/O operations during migration. The resulting increase in service latency could be very costly to a business. This paper presents a novel storage migration scheduling algorithm that can greatly improve storage I/O performance during wide-area migration. Our algorithm is unique in that it considers individual virtual machine's storage I/O workload such as temporal locality, spatial locality and popularity characteristics to compute an efficient data transfer schedule. Using a fully implemented system on KVM and a trace-driven framework, we show that our algorithm provides large performance benefits across a wide range of popular virtual machine workloads.

As IP telephony becomes a critical part of the enterprise network infrastructure, there is an increasing need for tools to help manage VoIP systems. In this paper, we introduce DA VINCI, a tool to manage VoIP call routing - the core of VoIP configurations that determines how calls are routed inside the enterprise or between the enterprise and external networks. We first establish the complexity involved in call routing design and configuration by studying characteristics of nine large enterprise VoIP networks. We then develop methodologies to improve and simplify call routing configurations by (i) identifying common design patterns, i.e., shared resource routing and local resource routing, so that administrators can better understand and reuse existing call routing designs, and (ii) removing unused and superfluous configurations from dial plan designs while retaining the correctness of operations. All of our techniques are incorporated in the DA VINCI tool that has been deployed in more than ten enterprise networks to help administrators streamline VoIP maintenance through extraction, analysis, simulation, and configuration of call routing.

In this paper, we tackle challenges in migrating enterprise services into hybrid cloud-based deployments, where enterprise operations are partly hosted on-premise and partly in the cloud. Such hybrid architectures enable enterprises to benefit from cloud-based architectures, while honoring application performance requirements, and privacy restrictions on what services may be migrated to the cloud. We make several contributions. First, we highlight the complexity inherent in enterprise applications today in terms of their multi-tiered nature, large number of application components, and interdependencies. Second, we have developed a model to explore the benefits of a hybrid migration approach. Our model takes into account enterprise-specific constraints, cost savings, and increased transaction delays and wide-area communication costs that may result from the migration. Evaluations based on real enterprise applications and Azure-based cloud deployments show the benefits of a hybrid migration approach, and the importance of planning which components to migrate. Third, we shed insight on security policies associated with enterprise applications in data centers. We articulate the importance of ensuring assurable reconfiguration of security policies as enterprise applications are migrated to the cloud. We present algorithms to achieve this goal, and demonstrate their efficacy on realistic migration scenarios.

Abstract — Hosts participating in overlay multicast applications have a wide range of heterogeneity in bandwidth and participation characteristics. In this paper, we highlight and show the need to systematically consider prioritization as a key criterion in the design of protocols for overlay multicast. We identify trade-offs in the design of prioritization heuristics in two important contexts. The first part of the paper considers prioritization strategies in the context of heterogeneity in node outgoing bandwidth and node stay time durations, and a lack of correlation between the two dimensions. The second part of the paper considers bandwidth allocation and prioritization policies with multi-tree data delivery in environments with heterogeneity in outgoing bandwidth and a certain degree of altruistic behavior. We conduct a systematic study of the trade-offs using both real trace data, and sensitivity studies using synthetic workloads. To the best of our knowledge, this is the first work to identify and study these trade-offs, and to demonstrate the potential benefits of the resulting prioritization heuristics. I.

In this paper, we study the live streaming workload from a large content delivery network. Our data, collected over a 3 month period, contains over 70 million requests for 5,000 distinct URLs from clients in over 200 countries. To our knowledge, this is the most extensive data of live streaming on the Internet that has been studied to date. Our contributions are two-fold. First, we present a macroscopic analysis of the workload, characterizing popularity, arrival process, session duration, and transport protocol use. Our results show that popularity follows a 2-mode Zipf distribution, session interarrivals within small time-windows are exponential, session durations are heavy-tailed, and that UDP is far from having universal reach on the Internet. Second, we cover two additional characteristics that are more specific to the nature of live streaming applications: the diversity of clients in comparison to traditional broadcast media like radio and TV, and the phenomena that many clients regularly join recurring events. We find that Internet streaming does reach a wide audience, often spanning hundreds of AS domains and tens of countries. More interesting is that small streams also have a diverse audience. We also find that recurring users often have lifetimes of at least as long as one-third of the days in the event.

While application end-point architectures have proven to be viable solutions for large-scale distributed applications such as distributed computing and file-sharing, there is little known about its feasibility for more bandwidth-demanding applications such as live streaming. Heterogeneity in bandwidth resources and dynamic group membership, inherent properties of application end-points, may adversely affect the construction of a usable and efficient overlay. At large scales, the problems become even more challenging. In this paper, we study one of the most prominent architectural issues in overlay multicast: the feasibility of supporting large-scale groups using an application end-point architecture. We look at three key requirements for feasibility: (i) are there enough resources to construct an overlay, (ii) can a stable and connected overlay be maintained in the presence of group dynamics, and (iii) can an efficient overlay be constructed? Using traces from a large content delivery network, we characterize the behavior of users watching live audio and video streams. We show that in many common real-world scenarios, all three requirements are satisfied. In addition, we evaluate the performance of several design alternatives and show that simple algorithms have the potential to meet these requirements in practice. Overall, our results argue for the feasibility of supporting large-scale live streaming using an application end-point architecture.

Measurement-based optimization is one important strategy to improve the performance of bandwidth-demanding peer-to-peer systems. However, to date, we have little quantitative knowledge of how well basic lightweight measurement-based techniques such as RTT probing, 10KB TCP probing, and bottleneck bandwidth probing may work in practice in the peer-to-peer environment. By conducting trace-based analyses, we find that the basic techniques can help achieve 40 to 50% optimal performance. To deepen our understanding, we analyze some of the intrinsic properties of these techniques. Our analyses reveal the inherent difficulty of the peer selection problem due to the extreme heterogeneity in the peer-to-peer environment, and that the basic techniques are limited because their primary strength lies in eliminating the low-performance peers rather than reliably identifying the best-performing one. However, our analyses also reveal two key insights that can potentially be exploited by applications. First, for adaptive applications that can continuously change communication peers, the basic techniques are highly effective in guiding the adaption process. In our experiments, typically an 80% optimal peer can be found by trying less than 5 candidates. Secondly, we find that the basic techniques are highly complementary and can potentially be combined to better identify a high-performance peer, thus even applications that cannot adapt may benefit. Using media file sharing and overlay multicast streaming as case studies, we have systematically experimented with several simple combined peer selection techniques. Our results show that for the nonadaptive media file sharing application, a simple combined technique can boost performance to 60% optimal. In contrast, for the continuously adaptive overlay multicast application, we find that a basic technique with even low-fidelity network information is sufficient to ensure good performance. We believe our findings will help guide the future designs of high-performance peer-to-peer systems.

Locating content in decentralized peer-to-peer systems is a challenging problem. Gnutella, a popular file-sharing application, relies on flooding queries to all peers. Although flooding is simple and robust, it is not scalable. We explore how to retain the simplicity of Gnutella, while addressing its inherent weakness: scalability. We propose a content location solution in which peers loosely organize themselves into an interest-based structure on top of the existing Gnutella network. Our approach exploits a simple, yet powerful principle called interest-based locality, which posits that if a peer has a particular piece of content that one is interested in, it is very likely that it will have other items that one is interested in as well. When using our algorithm, called interest-based shortcuts, a significant amount of flooding can be avoided, making Gnutella a more competitive solution. In addition, shortcuts are modular and can be used to improve the performance of other content location mechanisms including distributed hash table schemes. We demonstrate the existence of interest-based locality in five diverse traces of content distribution applications, two of which are traces of popular peer-to-peer file-sharing applications. Simulation results show that interest-based shortcuts often resolve queries quickly in one peer-to-peer hop, while reducing the total load in the system by a factor of 3 to 7.

In this paper, we discuss the problem of distributing streaming media content, both live and on-demand, to a large number of hosts in a scalable way. Our work is set in the context of the traditional client-server framework. Specifically, we consider the problem that arises when the server is overwhelmed by the volume of requests from its clients. As a solution, we propose Cooperative Networking (CoopNet), where clients cooperate to distribute content, thereby alleviating the load on the server. We discuss the proposed solution in some detail, pointing out the interesting research issues that arise, and present a preliminary evaluation using traces gathered at a busy news site during the ash crowd that occurred on September 11, 2001.

We propose a third-party value-added services framework for enhancing the performance of reliable multicast protocols.In this framework, a value-added service provider will place servers called waypoints throughout ISPs' networks.Waypoints run a fully distributed, dynamic algorithm to determine which multicast groups to join.Having joined a group, a waypoint participates in the error recovery protocol, supplying repairs to receivers.From the application's perspective, waypoints appear to be additional application endpoints in the network.Waypoints seamlessly interoperate with current reliable multicast algorithms with only a minor change to receivers and no changes to routers.In our implementation, receivers and waypoints use STORM, a structure-based error recovery protocol.The waypoint recovery service is not limited to one error recovery protocol, and can be extended to enhance other reliable multicast protocols.Results from simulation experiments are presented to evaluate the potential benefits of theproposed scheme.We find that when multicast group members are isolated from each other, a waypoint recovery service can significantly enhance receivers' performance.