The EdgeNet software is free, open-source, liberally licensed code that extends the Kubernetes container orchestration system to the edge cloud. We use this code to run the EdgeNet testbed, an internet-scale edge cloud for distributed systems researchers. This demonstration showcases three features of EdgeNet: its multitenancy model, its multi-provider aspect, and its geographically-based selective deployment capability. Multitenancy allows multiple teams to use the platform concurrently; being multi-provider, independent contributors can make nodes available to the platform; and selective deployment facilitates location-based placement of software. Under our guidance, demo participants invoke the Kubernetes command-line interface to use the testbed. In so doing, they get experience with the testbed, which they can continue to use afterwards. They also gain insight into how the demonstrated features are useful for edge cloud container deployment in general. Participants who volunteer to help EdgeNet nodes receive Odroid devices to host in their homes or workplaces.

EdgeNet is the prototype of a scalable, sustainable general-purpose testbed for very wide area distributed systems and extremely low-latency distributed services. In this, it is aimed at the same experimenters and systems that formed the core usage of previous, highly-successful wide-area testbeds such as PlanetLab [1], G-Lab [2], V-Node [5], GENI [4], and SAVI [3], and it incorporates many of the features that characterized those previous testbeds. EdgeNet's goal is to achieve the usability and research value of the previous generations of wide area testbed, whilst offering radical improvements in the scalability and sustainability of those systems. It achieves this scalability and sustainability through a strategy of using industry-standard open-source software as the basis of its software stack, and by a strategy of hardwarefree, bottom-up, site-driven deployment. EdgeNet follows the Seattle [6] and PlanetIgnite [9] strategy of permitting sites to join the testbed with purely local action.

Distributed platforms are now a de facto standard in modern software and application development. Although the ACM/IEEE Curriculum 2013 introduces Parallel and Distributed Computing as a first class knowledge area for the first time, the right level of abstraction to teach these concepts is still an important question that needs to be explored. This work presents our findings in teaching cloud computing by exposing upper-level students to testbeds in use by the distributed systems research community. The possibility of giving students practical and relevant experience was explored in the context of new course assignment objectives. Furthermore, students were able to significantly contribute to a pilot class project with medium-scale computation based on satellite data. However, the software engineering challenges in these environments proved to be daunting. In particular, these challenges were exacerbated by a lack of debugging support relative to the environments students were more familiar with---requiring development practices that out-stripped typical course experiences. Our proposed set of experiments and project provide a basis for an evaluation of the trade-offs of teaching cloud and distributed systems on the wild side. We hope that these findings provide insight into some of the possibilities to consider when preparing the next generation of computer scientists to engage with software practices and paradigms that are already fundamental in today's highly distributed systems.

In this paper, we consider a network of OpenFlow switches as an acyclic network of high-dimensional Boolean functions. We reduce classic network properties to logic functions over the variables of this network, and demonstrate that these properties hold if and only if the conjunction of the derived Boolean network and proposition is satisfied. We demonstrate that the derived satisfiability instance is polynomially related to the size of the switch network and the network property. The problem of verification of OpenFlow networks is thus demonstrated to be in the class NP. We show that OpenFlow Verification is NP-complete by a reduction from SAT. We further consider a slight restriction in the OpenFlow rule set to prefix rules, and demonstrate that OpenFlow Verification is polynomial when the ruleset is restricted to prefix rules.

Network device vendors have recently opened up the processing capabilities on their hardware platform to support third-party applications. In this paper, we explore the requirements and overheads associated with co-locating middlebox functionality on such computing resources on networking hardware. In particular, we use an example of TCP acceleration proxy (CHART) that improves throughput over networks with delay and loss. The CHART system, developed by HP and its partners provides enhanced TCP/IP performance and service quality guarantees by deploying performance accelerating proxies, which enables legacy clients to benefit by high-performance network service. Use of the TCP proxy, however, requires manual configuration on the clients changing http proxy and/or routing table settings. Can we remove the need to configure end-hosts by inserting a transparent TCP proxy in the path, without losing performance? To address this question, we implement the accelerator on HP's x86-based processing blade designed to integrate network applications within switch architecture as well as on low-end home routers with OpenWRT. We describe the implementation detail such as flow redirection for transparency and new mechanisms required for easy insertion of proxies in the network path. We also evaluate its performance on HP's experimental testbed in terms of throughput and additional processing overhead.

Reconfigurable network hardware makes it easier to experiment with and prototype high-speed networking systems. However, these devices are still relatively hard to program; for example, requiring users to develop in Verilog or VHDL. Further, these devices are commonly designed to work with software on a host computer, requiring the co-development of these hardware and software components.

In this paper, we consider migration of virtual machines in a data center to minimize network power consumption. Network power is consumed when switches are turned on, and conserved when they are turned off; the optimization problem then is to site virtual machines within the data center to achieve connectivity and desired bandwidth while turning on as few switches as possible. Depending upon specifics of the permissible optimization, and topology considerations within the data center, this optimization problem can be easy, or hard. We fully taxonomize the suite of optimization problems in this general space, showing that the most complex problem (placement and routing of virtual machines in a topologically-rich data center network) is NP- hard. We offer a placement technique based on a classic VLSI placement algorithm, and demonstrate efficacy on a trace set derived from a production data center.

The Internet was designed as a packet-switched network in the 1960's and 1970's, with the explicit intent of sacrificing quality-of-service guarantees for an individual application in order to optimize channel usage and provide optimal median service for all applications. This approach was successful, since the application mix of the Internet heretofore has been dominated by applications with low quality-of-service needs: primarily bulk data transfer and low-bandwidth text- based interactive applications. As the Internet absorbs other networks and applications with strong quality-of-service requirements (television, voice over IP) , this tradeoff changes. We are faced with the problem of introducing the quality- of-service guarantees of circuit-switching into packet-switched networks. Fundamental change to the lower layers of the Internet stack have proven infeasible; even strongly-motivated, well-designed modifications which made transition a first-class design consideration have had difficult introductions. ATM and IPv6 are two recent examples. One effective transition strategy for new networking techniques has been the use of overlays. In this paper, we introduce the concept of a service overlay network, to offer circuit-switched behavior on legacy IP networks, and establish the requirements on the underlying IP network to make this strategy effective.

Packet classification is a function increasingly used in a number of networking appliances and applications. Typically, this consists of a set of abstract classifications, and a set of rules which sort packets into the various classifications. For packet classification at line speeds, Ternary Content-Addressable Memories (TCAMs) have become a norm in most networking hardware. However, TCAMs are expensive and power-hungry. Hence, a packet classification ruleset need to be minimized before populating the TCAM. In this paper, we formulate the Ruleset Minimization Problem for TCAM as an abstract optimization problem based on two-level logic minimization, and propose an exact solution and a number of heuristics. We present experimental results with two different datasets-artificial filter sets generated using ClassBench tool suite and a real firewall Access Control List (ACL) from a large enterprise. We observe an average reduction of 41% in artificial filter sets and 72.5% reduction in the firewall ACL using the proposed heuristics.

To improve the performance of VPN connections we investigate how the bandwidth of multiple access links can be aggregated with inverse multiplexing to create a single, higher capacity logical communication link. But achieving the maximum possible aggregated TCP throughput becomes extremely challenging if the underlying links either use different technologies (e.g., DSL, cable modem) or suffer different or time-varying communication characteristics (e.g., available bandwidth, packet loss rate). To maximize VPN throughput we have constructed a system that combines two distinct innovations. First, we continuously measure the communication characteristics of the underlying component links in our aggregate and dynamically assign packets to each link in proportion to its available capacity. Second, we modify TCP congestion control across the inverse-multiplexed access hop to avoid rate decreases normally initiated by the delayed acknowledgments often triggered when using legacy TCP on multiple heterogeneous paths. We describe the systempsilas implementation, the test environment we built on Emulab, and show that when access links form the communication bottleneck in the end-to-end connection we can significantly increase VPN performance over conventional approaches.