Joshua Reich

Languages for software-defined networks

Modern computer networks perform a bewildering array of tasks, from routing and traffic monitoring, to access control and server load balancing. However, managing these networks is unnecessarily complicated and error-prone, due to a heterogeneous mix of devices (e.g., routers, switches, firewalls, and middleboxes) with closed and proprietary configuration interfaces. Softwaredefined networks are poised to change this by offering a clean and open interface between networking devices and the software that controls them. In particular, many commercial switches support the OpenFlow protocol, and a number of campus, data center, and backbone networks have deployed the new technology. However, while SDNs make it possible to program the network, they does not make it easy. Todays OpenFlow controllers offer low-level APIs that mimic the underlying switch hardware. To reach SDNs full potential, we need to identify the right higher-level abstractions for creating (and composing) applications. In the Frenetic project, we are designing simple and intuitive abstractions for programming the three main stages of network management: monitoring network traffic, specifying and composing packet forwarding policies, and updating policies in a consistent way. Overall, these abstractions make it dramatically easier for programmers to write and reason about SDN applications.

The age of impatience: optimal replication schemes for opportunistic networks

Multimedia content dissemination in mobile settings requires significant bandwidth. Centralized infrastructure is often either inadequate or overly expensive to fill the demand. Here, we study an alternative P2P content dissemination scheme for mobile devices (e.g., smart-phones), which leverages local dedicated caches on these devices to opportunistically fulfill user requests. In our model, the allocation of content in the global distributed cache comprising the union of all local caches, determines the pattern of demand fulfillment. By selectively replicating local content at node meetings, the global cache can be driven towards a more efficient allocation. However, the allocations efficiency itself is determined by a previously overlooked factor - the impatience of content requesters. By describing user impatience in the form of any monotonically decreasing delay-utility functions, we show that an optimal allocation can be efficient computed or approximated. As users become increasingly impatient, the optimal allocation varies steadily between uniform and highly-skewed towards popular content. Moreover, in opportunistic environments, the global cache state may be difficult or impossible to obtain, requiring that replication decisions be made using only local knowledge. We develop a reactive distributed algorithm, Query Counting Replication (QCR) that for any delay-utility function drives the global cache towards the optimal allocation - without use of any explicit estimators or control channel information. We validate our techniques on real-world contact traces, demonstrating the robustness of our analytic results in the face of heterogeneous meeting rates and bursty contacts. We find QCR compares favorably to a variety of heuristic perfect control-channel competitors.

HotSwap: correct and efficient controller upgrades for software-defined networks

Like any complex software, SDN programs must be updated periodically, whether to migrate to a new controller platform, repair bugs, or address performance issues. Nowadays, SDN operators typically perform such upgrades by stopping the old controller and starting the new one---an approach that wipes out all installed flow table entries and causes substantial disruption including losing packets, increasing latency, and even compromising correctness. This paper presents HotSwap, a system for upgrading SDN controllers in a disruption-free and correct manner. HotSwap is a hypervisor (sitting between the switches and the controller) that maintains a history of network events. To upgrade from an old controller to a new one, HotSwap bootstraps the new controller (by replaying the history) and monitors its output (to determine which parts of the network state may be reused with the new controller). To ensure good performance, HotSwap filters the history using queries specified by programmers. We describe our design and preliminary implementation of HotSwap, and present experimental results demonstrating its effectiveness for managing upgrades to third-party controller programs.

Connectivity Maintenance in Mobile Wireless Networks via Constrained Mobility

We explore distributed mechanisms for maintaining the physical layer connectivity of a mobile wireless network while still permitting significant area coverage. Moreover, we require that these mechanisms maintain connectivity despite the unpredictable wireless propagation behavior found in complex real-world environments. To this end, we propose the Spreadable Connected Autonomic Network (SCAN) algorithm, a fully distributed, on-line, low overhead mechanism for maintaining the connectivity of a mobile wireless network. SCAN leverages knowledge of the local (2-hop) network topology to enable each node to intelligently halt its own movement and thereby avoid network partitioning events. By relying on topology data instead of locality information and deterministic connectivity models, SCAN can be applied in a wide range of realistic operational environments. We believe it is for precisely this reason that, to our best knowledge, SCAN was the first such approach to be implemented in hardware. Here, we present results from our implementation of SCAN, finding that our mobile robotic testbed maintains full connectivity over 99% of the time. Moreover, SCAN achieves this in a complex indoor environment, while still allowing testbed nodes to cover a significant area.

VMTorrent: scalable P2P virtual machine streaming

Clouds commonly store Virtual Machine (VM) images on networked storage. This poses a serious potential scalability bottleneck as launching a single fresh VM instance requires, at minimum, several hundred MB of network reads. As this bottleneck occurs most severely during read-intensive launching of new VMs, we focus on scalably minimizing time to boot a VM and load its critical applications. While effective scalable P2P streaming techniques for Video on Demand (VOD) scenarios where blocks arrive in-order and at constant rate are available, no techniques address scalable large-executable streaming. VM execution is non-deterministic, divergent, variable rate, and cannot miss blocks. VMTORRENT introduces a novel combination of block prioritization, profile-based execution prefetch, on-demand fetch, and decoupling of VM image presentation from underlying data-stream. VMTORRENT provides the first complete and effective solution to this growing scalability problem that is based on making better use of existing capacity, instead of throwing more hardware at it. Supported by analytic modeling, we present comprehensive experimental evaluation of VMTORRENT on real systems at scale, demonstrating the effectiveness of VMTORRENT. We find that VMTORRENT supports comparable execution time to that achieved using local disk. VMTORRENT maintains this performance while scaling to 100 instances, providing up to 11x speedup over current state-of-the-art and 30x over traditional network storage.

Roomba MADNeT: a mobile ad-hoc delay tolerant network testbed

We have built a mobile, ad-hoc, delay tolerant, network testbed (MADNeT). Our testbed is geared towards enabling the exploration of highly disconnected networks whose nodes must store and forward information (multi-hop paths are very unlikely), while allowing us to explore connected network scenarios as well. Using our MADNeT, we can emulate a wide range of mobile networking scenarios and execute real-world (albeit somewhat simplified) data collection missions. Our system is built from commercial off-the-shelf (COTS) hardware running a paired-down Linux OS, yielding a relatively inexpensive but fairly flexible mobile network testbed. We have implemented replication-based information diffusion protocols and a distributed opportunistic surveillance application on the testbed as proof-of-concept. Currently, we are refining our software tool-set and instrumenting our nodes to test both our protocols and application.

How green is IP-telephony?

With constantly increasing costs of energy, we ask ourselves what we can say about the energy efficiency of existing VoIP systems. To answer that question, we gather information about the existing client-server and peer-to-peer VoIP systems, build energy models for these systems, and evaluate their power consumption and relative energy efficiency through analysis and a series of experiments. Contrary to the recent work on energy efficiency of peer-to-peer systems, we find that even with efficient peers a peer-to-peer architecture can be less energy efficient than a client-server architecture. We also find that the presence of NATs in the network is a major obstacle in building energy efficient VoIP systems. We then provide a number of recommendations for making VoIP systems more energy efficient.

Connectivity maintenance in mobile wireless networks via constrained mobility

We explore distributed mechanisms for maintaining the physical layer connectivity of a mobile wireless network while still permitting significant area coverage. Moreover, we require that these mechanisms maintain connectivity despite the unpredictable wireless propagation behavior found in complex real-world environments. To this end, we propose the Spreadable Connected Autonomic Network (SCAN) algorithm, a fully distributed, on-line, low overhead mechanism for maintaining the connectivity of a mobile wireless network. SCAN leverages knowledge of the local (2-hop) network topology to enable each node to intelligently halt its own movement and thereby avoid network partitioning events. By relying on topology data instead of locality information and deterministic connectivity models, SCAN can be applied in a wide range of realistic operational environments. We believe it is for precisely this reason that, to our best knowledge, SCAN was the first such approach to be implemented in hardware. Here, we present results from our implementation of SCAN, finding that our mobile robotic testbed maintains full connectivity over 99% of the time. Moreover, SCAN achieves this in a complex indoor environment, while still allowing testbed nodes to cover a significant area.

Policy transformation in software defined networks

A Software Defined Network (SDN) enforces network-wide policies by installing packet-handling rules across a distributed collection of switches. Todays SDN platforms force programmers to decide how to decompose a high-level policy into the low-level rules in each switch. We argue that future SDN platforms should support automatic transformation of policies by moving, merging, or splitting rules across multiple switches. This would simplify programming by allowing programs written on one abstract switch to run over a more complex network topology, and simplify analysis by consolidating a policy spread over multiple switches into a single list of rules. This poster presents our ongoing work on a sound and complete set of axioms for policy transformation, to enable rewriting of rules across multiple switches while preserving the forwarding policy. These axioms are invaluable for creating and analyzing algorithms for optimizing the rewriting of rules.

Spreadable Connected Autonomic Networks (SCAN)

A Spreadable Connected Autonomic Network (SCAN) is a mobile network that automatically maintains its own connectivity as nodes move. We envision SCANs to enable a diverse set of applications such as self-spreading mesh networks and robotic search and rescue systems. This paper describes our experiences developing a prototype robotic SCAN built from commercial, off-the-shelf hardware, to support such applications. A major contribution of our work is the development of a protocol, called SCAN1, which maintains network connectivity by enabling individual nodes to determine when they must constrain their mobility in order to avoid disconnecting the network. SCAN1 achieves its goal through an entirely distributed process in which individual nodes utilize only local (2-hop) knowledge of the network’s topology to periodically make a simple decision: move, or freeze in place. Along with experimental results from our hardware testbed, wemodel SCAN1’s performance, providing both supporting analysis and simulation for the efficacy of SCAN1 as a solution to enable SCANs. While our evaluation of SCAN1 in this paper is limited to systems whose capabilities match those of our testbed, SCAN1 can be utilized in conjunction with a wide-range of potential applications and environments, as either a primary or backup connectivity maintenance mechanism.