Pramod V. Koppol

Incremental integration testing of concurrent programs

We present a method for selecting test sequences for concurrent programs from labeled transitions systems (LTS). A common approach to selecting test sequences from a set of LTSs is to derive a global LTS, called the reachability graph, and then force deterministic program executions according to paths selected from the graph. However, using a reachability graph for test path selection introduces a state explosion problem. To overcome this problem, a reduced graph can be generated using incremental reachability analysis, which consists of repeatedly generating a reachability graph for a subset of LTSs, reducing this graph, and using the reduced graph in place of the original LTSs. Unfortunately, existing incremental reachability analysis techniques generate reduced graphs with insufficient information for deterministic testing. We present an incremental approach to testing concurrent programs. Incremental testing consists of incremental reachability analysis for test path selection and deterministic testing for test execution. We define a new type of reachability graph for incremental analysis, called an annotated labeled transition system (ALTS). An ALTS is an LTS annotated with information necessary for deterministic testing. We propose practical coverage criteria for selecting tests paths from an ALTS and present an ALTS reduction algorithm. The results of several case studies are reported.

Improving Zap Response Time for IPTV

Channel zapping is the act of changing from one television channel to another. Zap response time is the time it takes for the new TV channel to start playing from the time a request to zap to that channel occurs. In digital TV systems such as IPTV where TV channel content is transported as an MPEG-2 or MPEG-4 data stream encapsulated in IP packets, zap response time is a significant concern. This is due to the requirement that for an MPEG stream, which is a series of groups of pictures (GOPs), playout can only start at the beginning of a GOP. The time between the occurrence of a zap event to the time to the beginning of the next GOP, which can be anywhere from less than a second to a few seconds, is a significant contributor to zap response time. It is widely accepted that such large zap response times can significantly hamper IPTV quality of experience (QoE) and an efficient solution to this problem is therefore a business imperative. Existing solutions to this problem are bandwidth inefficient either on the last mile, or within the video distribution network, or both. We present a novel solution that is significantly efficient in terms of bandwidth utilization and zap response time. We derive and present the theoretical bounds for our solution and quantitatively demonstrate the properties of our solution through results from extensive simulations.

An Internet scale simulation setup for BGP

BGP convergence issues in the Internet have been extensively studied in recent years. It has been shown through formal analysis that BGP convergence is not always guaranteed. When it does converge, active measurement based techniques have determined that the convergence process may take upto 15 minutes while also causing intermittent packet loss/latency. As BGP enhancements are explored to address this convergence problem, controlled simulation experiments offer the most convenient and feasible mechanism to compare the efficacy of enhanced BGP procedures with respect to the current protocol mechanisms. Since the convergence behavior can be significantly influenced by network topology, it is desirable to use large Internet like topologies in such controlled simulations.In this paper, we present first steps towards realizing a large scale BGP simulation environment for reasoning about current BGP procedures and also for evaluating the efficacy of BGP protocol enhancements.

Optimal Construction of Redundant Multicast Trees in Directed Graphs

In this paper, we consider the problem of protection of multicast connections against single link or node failures using redundant multicast trees (RMTs). RMTs connect the source node of a multicast connection to all its destinations in such a way that in the event of a single link or node failure in the network, every destination node is still connected to the source node in at least one of the two trees. Construction of RMTs has been studied extensively for undirected graphs, but not for directed graphs. Computing RMTs in directed graphs can be important in networks, for instance, where link capacity is available in one direction but not the other. Also, none of the schemes proposed in previous work provide a solution for finding optimal (minimal-cost) RMTs even for weighted undirected graphs. We show that a whole class of earlier schemes that are based on the concept of ear-decomposition cannot find optimal redundant trees in certain network topologies. To our knowledge, we are the first to consider the problem of finding optimal RMTs in directed graphs. We present a novel efficient scheme for construction of optimal RMTs in networks modeled as weighted undirected as well as directed graphs.

A simple IP fast reroute scheme for full coverage

A major concern in IP networks is to ensure that any topology changes, whether planned or unplanned, do not disrupt network performance. IP Fast Reroute (IP FRR) is a general approach to address this issue, by promptly forwarding an IP packet to a predetermined alternate next hop as soon as the primary next hop to the destination becomes unavailable. Among the numerous IP FRR schemes proposed to date, the simplest ones do not guarantee protection against every component failure, while more sophisticated ones tend to be difficult to implement due to various inherent complexities, such as nontrivial modifications of IP packets or high resource requirements.

Efficient construction of directed Redundant Steiner trees

Multicast is getting increasingly popular to support critical business services such as IPTV, video conferencing and content distribution. To meet the stringent availability requirements, a highly fault resilient multicast infrastructure is critically important. Redundant Steiner trees (RST) are one mechanism for realizing such infrastructure. RSTs are a pair of Steiner trees that ensure that for any single link or node failure in the network, every destination node of a multicast connection is reachable from the source on at least one of these two trees. In this paper, we consider the problem of constructing optimal RSTs in a directed graph, which is NP-complete. In light of this, we propose a simple and efficient scheme to construct near-optimal RSTs. The scheme ensures an approximation ratio of 2k for general directed graphs, where k denotes the number of destination nodes. We compare the performance of our scheme with an linear program (LP) formulation of the problem that provides a lower bound on the optimal solution. Our simulations illustrate that our scheme constructs near optimal RSTs independent of the number of destination nodes and significantly outperforms other approaches proposed in prior literature.

Scalable and elastic telecommunication services in the cloud

We consider the problem of virtualization of telecommunication (telecom) services, which are session-oriented by nature and currently run on dedicated hardware. Such virtualization involves eliminating the dedicated hardware and moving the software to run in virtual machines within a standard cloud computing infrastructure. Existing mechanisms for running services in the cloud are typically targeted towards web services with short-lived sessions where the services are often resilient to intermittent failures of the network or endpoints. We claim that such mechanisms are inadequate for implementations of typical telecom services where sessions are often long-lived and service users are intolerant of service and network interruptions. We present a framework, which we refer to as Telecom as a Service (TaaS), and the associated TaaS primitives for realizing highly scalable and elastic session-oriented telecommunications services in the cloud while imposing minimal changes to existing services and no changes to client implementations. We show through empirical data that our mechanisms scale with network load.

Resource reservation comparison of fault resilient routing schemes

Reliable delivery of network services is critically for numerous real-time applications, such as video conferencing, broadcast TV and content distribution, in which time sensitive content is delivered to a single or multiple destinations. Failure protection in connection-oriented networks can be realized using local protection schemes, such as fast reroute (FRR), or using end-to-end protection schemes, like redundant trees. In this paper we study the trade-offs between local and end-to-end protection schemes from several key aspects such as protection availability, management complexity and in particular the resource reservation efficiency of the two approaches.

The inherent difficulty of timely primary-backup replication

We show that existing methods for primary-backup replication may disrupt the timing behavior of an underlying service to the extent of making it unusable. We prove that the problem is inherent to the primary-backup model.

Link-coloring based scheme for multicast and unicast protection

We propose a computationally efficient link-coloring based scheme that facilitates end-to-end protection for dynamic multicast and unicast connections. For a given source node, links are colored as red or blue such that, for any destination node, the red and blue paths are guaranteed to be node disjoint. Red and blue trees, constructed using only the red and blue links, respectively, always form redundant trees (RTs) independent of the used tree selection method. RTs connect the source of a multicast connection to all its destinations in such a way that in the event of a single link or node failure in the network, every destination is still connected to the source in at least one of the two trees. In network topologies where such RTs cannot be constructed due to lack of required path diversity, our scheme constructs two multicast trees such that the red and blue paths to any destination node only share cut links or nodes. To our knowledge, our scheme is the first that can construct such RTs. Although finding the optimal RTs is known to be NP-complete, extensive simulations show that our scheme calculates near optimal RTs and substantially outperforms other solutions.