Partnerships between academic institutions and industry for internships can benefit both parties. Experiential learning may help students prepare for their futures, while employers can help identify potential talent for jobs. Although the student perspective has been well explored, scholarship around employers remains limited. In this experience report, we describe a micro-internship program for (n=95) computing students that combined 7 weeks of university-led upskilling workshops with three weeks of practical experience with (n=18) employers to complete challenge projects in small groups. We elaborate further on the preparation and implementation required, as well as detail our qualitative evaluation of the program from the employer perspective. Situated learning theory (SLT) guided the investigation, which involved gathering feedback from employers through semi-structured interviews. Applying reflexive thematic analysis to employer interviews, we inductively examined: (1) how industry mentors integrated students into professional computing communities of practice (COP), (2) the mechanisms they employed to facilitate students' legitimate peripheral participation, (3) their observations of growth in students' technical and professional skills and identity formation, and (4) reciprocal learning that occurred during the internship period. Six resulting themes were then deductively mapped to SLT sub-constructs to further understand employers' impact during computing internships. Employers facilitated authentic problem solving and real-world learning experiences for students while gaining new perspectives and insights regarding new technology and problem-solving approaches in the process.

Making the transition from academia to industry can be a rite of passage for college graduates. Internships allow students to gain experience in small doses, and help shape their career goals, actions, and decisions. We sought to explore how employers perceived undergraduate computing students' performance and experience in a three-week micro-internship. We applied the boundary crossing (BC) framework and analyzed and interpreted n = 49 quotes extracted from semi-structured interviews with three industry mentors using the methodology of framework analysis. We examined the quotes and categorized them into one of four mechanisms of BC: identification, coordination, reflection, and transformation. The greatest number of BC mechanisms reported was that of coordination at the interpersonal level (33%), where the interns interacted with their mentors to navigate the differences in expectations and tasks that they had already identified. 51% of the BC mechanisms were reported to be at the interpersonal level, while six instances of transformation at the institutional level were also observed in the analysis. Our study's results can help administrators and industry mentors gain insight into how computing students may leverage mentorship to navigate professional dynamics.

Scientific research is increasingly relying on complex workflows that span multiple computing paradigms, including high-performance computing (HPC), high-throughput computing (HTC), and machine learning/artificial intelligence (ML/AI). Traditional monolithic computing infrastructures often struggle to accommodate these diverse and evolving demands. The Reconfigurable Advanced Platform for Transdisciplinary Open Research (RAPTOR) addresses this challenge by providing a dynamically reconfigurable computing environment that integrates with federated resources. RAPTOR's architecture enables dynamic provisioning between an HPC cluster and the Chameleon Cloud platform based on workload requirements, supporting bare-metal customization for specialized applications. This paper focuses on RAPTOR's reconfigurability features and demonstrates their effectiveness through quantitative performance evaluations across four scientific domains: computational proteomics, climate modeling, weather research, and hurricane risk assessment. Our results demonstrate that RAPTOR's reconfigurable design significantly enhances research productivity by providing an appropriate computing environment for diverse computational needs.

The expanding applications of technology across sectors, coupled with the rising demand for qualified graduates, necessitate consideration of new ways to increase engagement with the discipline of computing. Towards this goal, we established a week-long program for non-majors to explore computing concepts (e.g., artificial intelligence) and aid in their professional development (e.g., through fostering presentation skills). We also sought to cultivate a community and incorporated peer and industry mentorship. In the experience report that follows, we detail the novel program and its evolution over five iterations across two institutions. We applied the Community of Inquiry framework to contextualize the programmatic design and its evaluation. Surveys collected daily gave insight into the student perspective on the various lessons and activities offered, with feedback from up to n = 141 students in total. Apart from including Likert-scale ratings to quantify preferences for each session, open-ended responses allowed greater understanding around what may have been viewed favorably or what could require further improvements. Based on the findings, we highlight how aspects of the experience may have contributed to the participants' engagement with the content, with others involved in the program, and with respect to learning outcomes. The session details and reflections presented are intended to inform as well as offer inspiration to other educators and administrators who may seek to introduce students from other majors to computing.

Internships can allow computing students to cultivate valuable skills while offering them practical insight into industry. The aim of our study was to gain an understanding of undergraduate computing students' perceptions of a three-week micro-internship (called a ''Sprinternship'') program. We sought to explore their experiences throughout its duration, which included a priori professional and technical development training. We applied the methodology of phenomenography, conducting semi-structured interviews with n = 27 students and taking the developmental approach to the analysis. We noted cognitive, affective, interpersonal, and career-oriented factors often influenced students' views of the experience. In this work, we share the seven categories of description that emerged from the analysis and provide the implications. The findings of this investigation can offer guidance for educators and administrators looking to create similar short-term internship opportunities.

Cache management is a critical aspect of computer architecture, encompassing techniques such as cache replacement, bypassing, and prefetching. Existing research has often focused on individual techniques, overlooking the potential benefits of joint optimization. Moreover, many of these approaches rely on static and intuition-driven policies, limiting their performance under complex and dynamic workloads. To address these challenges, this paper introduces CHROME, a novel concurrencyaware cache management framework. CHROME takes a holistic approach by seamlessly integrating intelligent cache replacement and bypassing with pattern-based prefetching. By leveraging online reinforcement learning, CHROME dynamically adapts cache decisions based on multiple program features and applies a reward for each decision that considers the accuracy of the action and the system-level feedback information. Our performance evaluation demonstrates that CHROME outperforms current state-of-the-art schemes, exhibiting significant improvements in cache management. Notably, CHROME achieves a remarkable performance boost of up to 13.7% over the traditional LRU method in multi-core systems with only modest overhead.

The demand for skilled workers in computing continues to outpace the supply of qualified graduates. Despite the need, hiring can be challenging, both for employers seeking prospective employees and for students who may be unsure where to apply, daunted by technical interviews, and/or feeling the effects of imposter phenomena. In this experience report, we describe a program established to reduce some of these hurdles by pairing (n = 63) undergraduate students with (n = 7) companies to offer short-term computing internships, called a Sprinternship. Sprinternships eliminated the hurdle of technical interviews, provided students with training beforehand to offer foundational knowledge, and placed them in teams to work on challenge projects. We describe the details of the program and our investigation of its impact. Social Cognitive Career Theory guided the inquiry as we took a mixed-methods approach to understand the students' experiences and the potential impact on their self-efficacy, outcome expectations, and career goals. Quantitative analysis revealed a statistically significant increase in students' confidence in computing, something echoed in their open-ended responses. Thematic analysis further yielded that Sprinternships were meaningful in two major areas: Goals and Learning Experiences. The program aided in students' self-discovery, made them feel accomplished, and strengthened their industry ambitions. Responses also revealed positive and negative programmatic aspects to consider for future iterations. We hope that our description of the Sprinternships, findings, and recommendations can be useful to other practitioners looking to engage students with practical learning and enhance their graduate employability.

In recent years, exploring autonomous vehicles has become an emerging research topic due to the massive opportuni-ties to deploy deep learning models in real-world transportation applications. Although deep learning models have been widely applied to microscopic traffic simulations (MTS), they are limited to offline prescriptive analysis or predictive analysis. However, dynamics of a traffic environment frequently change over time; little research has been focused on dynamically updating deep learning models parameters in response to real-time traffic conditions. Further, for dynamic model parameter updates, it is difficult to preserve privacy since raw data is transmitted to the server for model updating (training). In this paper, we present a federated multimodal microscopic traffic simulation (FedMMTS) framework that incorporates federated learning to preserve data privacy and applies multimodal data analytics to achieve a more accurate sensing of the environment. We present preliminary experiments to demonstrate the effectiveness of this framework in a microscopic traffic environment. To the best of our knowledge, this is the first work that incorporates federated learning and data multimodality in MTS environments.

As software programs, Virtual Network Functions (VNFs) introduce new challenges to network availability due to their potential software failures. Existing models on the availability of VNF networks did not consider all the possible hardware and software failures, making them incapable of analyzing the end-to-end availability of a path. Furthermore, they did not capture the correlation between repeating nodes and links in the path, resulting in inaccurate analytical results. In this paper, we propose a new analytical model, which considers all hardware and software failures as well as the effect of repeating components, to effectively analyze the end-to-end availability of a flow path in VNF networks. On top of the analytical model, we formulate the Highest Availability Path (HAP) problem that finds the flow path with the highest end-to-end availability, and prove its NP-hardness by reduction from the Node-Weighted Steiner Tree problem. Next, we propose two algorithms for HAP: the first one based on a Steiner Tree approximation algorithm, having high time complexity and serving as a performance benchmark; the second one using a dynamic programming approach to search a multi-layer graph in polynomial time. Finally, we present extensive evaluation data to demonstrate the effectiveness of the Layered Search algorithm, which achieves comparable performance as that of the Steiner Tree based algorithm and runs faster by four orders of magnitude.

Data centers today host large numbers of workloads and many of these workloads consume significant storage resources. Given the long history of successes in storage caching, it is only natural such successes bear fruit in modern data centers, at scale. This paper presents CaaS, a generalized caching service for cloud data centers. Departing from existing application, storage, or data-type specific caches, CaaS unifies and abstracts data center caching resources making these available to any workload and for any data type. Also departing from past caching practices, CaaS is fault-tolerant allowing it to cache writes without risk of data loss. We expect that systems such as CaaS will help bridge the gap between heterogeneous and distributed cache resources and data-intensive applications in a data center.

Material screening entails a large number of electronic structure simulations. Traditionally, these simulation runs are treated separately as solving independent Kohn-Sham (KS) equations. In this paper, we formulate material screening as an inter-job scheduling problem for solving a system of KS equations, and in doing so allowing one to explore different scheduling methods that use the results of some equations to expedite the solution of others. We propose the concept of sharing iterative simulation and employ several optimization methods to initialize a simulation run using the distribution of particles from similar jobs as the initial condition. More specifically, we propose two similarity metrics, one qualitative and the other quantitative, to predict the simulation runtime of a material screen job based on its similarity to other jobs. Accordingly, we present two inter-job scheduling algorithms that make use the qualitative and quantitative similarity information. We conducted extensive experiments on the Sunway TaihuLight supercomputer for a practical material screening problem to evaluate the performance of the two scheduling algorithms using the proposed similarity metrics. We show that the total time required to run the large number of material screening jobs can be significantly reduced, and the algorithms are robust even with moderate inaccurate prediction on the simulation runtime. The quantitative algorithm achieves better results than the qualitative algorithm using more accurate prediction and thus achieving more significant runtime reduction.

Reproducing network emulation experiments on diverse physical platforms with varying computation and communication resources is non-trivial. Many state-of-the-art network emulation testbeds do not guarantee timing fidelity. Consequently, results obtained from these testbeds can be misleading, especially when insufficient physical resources are provided to run the experiments. Reproducibility is far from being the norm. In this paper, we present a novel approach that can guarantee reproducible results for network emulation. Our system, called the Virtual Time Machine (VTM), takes advantage of both time dilation and carefully controlled scheduling of the virtual machines. Time dilation allows sufficiently scaled resources to run the experiments in virtual time, and controlled VM scheduling prescribes the precise timing of message passing for distributed applications---independent of the resource provisioning of the underlying physical testbed. Preliminary experiments show that VTM can guarantee reproducible results with varying time dilation, resource subscription, and VM scheduling scenarios.

There has been a dramatic increase in the popularity of Container as a Service (CaaS) clouds. The CaaS multi-tier applications could be optimized by using network topology, link or server load knowledge to choose the best endpoints to run in CaaS cloud. However, it is difficult to apply those optimizations to the public datacenter shared by multi-tenants. This is because of the opacity between the tenants and the datacenter providers: Providers have no insight into tenant's container workloads and dependencies, while tenants have no clue about the underlying network topology, link, and load. As a result, containers might be booted at wrong physical nodes that lead to performance degradation due to bi-section bandwidth bottleneck or co-located container interference. We propose 'DocMan', a toolset that adopts a black-box approach to discover container ensembles and collect information about intra-ensemble container interactions. It uses a combination of techniques such as distance identification and hierarchical clustering. The experimental results demonstrate that DocMan enables optimized containers placement to reduce the stress on bi-section bandwidth of the datacenter's network. The method can detect container ensembles at low cost and with 92% accuracy and significantly improve performance for multi-tier applications under the best of circumstances.

Parallel application performance models provide valuable insight about the performance in real systems. Capable tools providing fast, accurate, and comprehensive prediction and evaluation of high-performance computing (HPC) applications and system architectures have important value. This paper presents PyPassT, an analysis based modeling framework built on static program analysis and integrated simulation of the target HPC architectures. More specifically, the framework analyzes application source code written in C with OpenACC directives and transforms it into an application model describing its computation and communication behavior (including CPU and GPU workloads, memory accesses, and message-passing transactions). The application model is then executed on a simulated HPC architecture for performance analysis. Preliminary experiments demonstrate that the proposed framework can represent the runtime behavior of benchmark applications with good accuracy.

Mininet is a container-based emulation environment that can study networks with virtual hosts and OpenFlow-enabled virtual switches on Linux. However, it is well-known that experiments using Mininet may lose fidelity for large-scale networks and heavy traffic load. One solution is to use a distributed setup where an experiment constitutes multiple instances of Mininet running on a cluster, each handling a subset of virtual hosts and switches. Such arrangement, however, is still constrained by bandwidth and latency limitations in the physical connection between the instances. In this paper, we propose a novel method of integrating distributed Mininet instances using a symbiotic approach, which extends an existing method for combining real-time simulation and emulation. We use an abstract network model to coordinate the distributed instances, which are superimposed to represent the target network. In this case, one can more effectively study the behavior of real implementation of network applications on large-scale networks, since the interaction between the Mininet instances is only capturing the effect of contentions among network flows in shared queues, as opposed to having to exchange individual network packets, which can be limited by bandwidth or sensitive to latency. We provide a prototype implementation of the new approach and present validation studies to show it can achieve accurate results. We also present a case study that successfully replicates the behavior of a denial-of-service (DoS) attack protocol.

Demand response refers to reducing energy consumption of participating systems in response to transient surge in power demand or other emergency events. Demand response is particularly important for maintaining power grid transmission stability, as well as achieving overall energy saving. High Performance Computing (HPC) systems can be considered as ideal participants for demand-response programs, due to their massive energy demand. However, the potential loss of performance must be weighed against the possible gain in power system stability and energy reduction. In this paper, we explore the opportunity of demand response on HPC systems by proposing a new HPC job scheduling and resource provisioning model. More specifically, the proposed model applies power-bound energy-conservation job scheduling during the critical demand-response events, while maintaining the traditional performance-optimized job scheduling during the normal period. We expect such a model can attract willing participation of the HPC systems in the demand response programs, as it can improve both power stability and energy saving without significantly compromising application performance. We implement the proposed method in a simulator and compare it with the traditional scheduling approach. Using trace-driven simulation, we demonstrate that the HPC demand response is a viable approach toward power stability and energy savings with only marginal increase in the jobs' execution time.

Interconnection network is a critical component of high-performance computing architecture and application co-design. For many scientific applications, the increasing communication complexity poses a serious concern as it may hinder the scaling properties of these applications on novel architectures. It is apparent that a scalable, efficient, and accurate interconnect model would be essential for performance evaluation studies. In this paper, we present an interconnect model for predicting the performance of large-scale applications on high-performance architectures. In particular, we present a sufficiently detailed interconnect model for Cray's Gemini 3-D torus network. The model has been integrated with an implementation of the Message-Passing Interface (MPI) that can mimic most of its functions with packet-level accuracy on the target platform. Extensive experiments show that our integrated model provides good accuracy for predicting the network behavior, while at the same time allowing for good parallel scaling performance.

Mininet is a popular container-based emulation environment built on Linux for testing Open Flow applications. Using Mininet, one can compose an experimental network using a set of virtual hosts and virtual switches with flexibility. However, it is well understood that Mininet can only provide a limited capacity, both for CPU and network I/O, due to its underlying physical constraints. We propose a method for combining simulation and emulation to improve the scalability of network experiments. This is achieved by applying the symbiotic approach to effectively integrate emulation and simulation for hybrid experimentation. In this case, one can use Mininet to directly run Open Flow applications on the virtual machines and software switches, with network connectivity represented by detailed simulation at scale.

Large-scale network simulation imposes extremely high computing demand. While parallel processing techniques allows network simulation to scale up and benefit from contemporary high-end computing platforms, multi-resolutional modeling techniques, which differentiate network traffic representations in network models, can substantially reduce the computational requirement. In this paper, we present a novel method for offloading computationally intensive bulk traffic calculations to the background onto GPU, while leaving CPU to simulate detailed network transactions in the foreground. We present a hybrid traffic model that combines the foreground packet-oriented discrete-event simulation on CPU with the background fluid-based numerical calculations on GPU. In particular, we present several optimizations to efficiently integrate packet and fluid flows in simulation with overlapping computations on CPU and GPU. These optimizations exploit the lookahead inherent to the fluid equations, and take advantage of batch runs with fix-up computation and on-demand prefetching to reduce the frequency of interactions between CPU and GPU. Experiments show that our GPU-assisted hybrid traffic model can achieve substantial performance improvement over the CPU-only approach, while still maintaining good accuracy.

Data centers consume significant amounts of energy. As severs become more energy efficient with various energy saving techniques, the data center network (DCN) has been accounting for 20% or more of the energy consumed by the entire data center. While DCNs are typically provisioned with full bisection bandwidth, DCN traffic demonstrates fluctuating patterns. The objective of this work is to improve the energy efficiency of DCNs during off-peak traffic time by powering off idle devices. Although there exist a number of energy optimization solutions for DCNs, they consider only either the hosts or network, but not both. In this paper, we propose a joint optimization scheme that simultaneously optimizes virtual machine (VM) placement and network flow routing to maximize energy savings, and we also build an OpenFlow based prototype to experimentally demonstrate the effectiveness of our design. First, we formulate the joint optimization problem as an integer linear program, but it is not a practical solution due to high complexity. To practically and effectively combine host and network based optimization, we present a unified representation method that converts the VM placement problem to a routing problem. In addition, to accelerate processing the large number of servers and an even larger number of VMs, we describe a parallelization approach that divides the DCN into clusters for parallel processing. Further, to quickly find efficient paths for flows, we propose a fast topology oriented multipath routing algorithm that uses depth-first search to quickly traverse between hierarchical switch layers and uses the best-fit criterion to maximize flow consolidation. Finally, we have conducted extensive simulations and experiments to compare our design with existing ones. The simulation and experiment results fully demonstrate that our design outperforms existing hostor network-only optimization solutions, and well approximates the ideal linear program.

A testbed capable of representing detailed operations of complex applications under diverse large-scale network conditions can be extremely helpful for investigating potential system design and implementation problems, and studying application performance issues, such as scalability and robustness, even before the applications are deployed in a real environment. We introduce a novel method that combines high-performance large-scale network simulation and high-fidelity network emulation, and thereby enables real instances of network applications and protocols to run in real operating environments, and be tested under large-scale simulated network settings. In our approach, network simulation and emulation form a symbiotic relationship, through which they are synchronized for an accurate representation of the large-scale traffic behavior. We introduce a model downscaling method, along with an efficient queuing model and a traffic reproduction technique, which can significantly reduce the synchronization overhead and improve computational efficiency, while maintaining the accuracy of the system. We validate our approach with extensive experiments via simulation and with a real-system prototype.

Flow-level bandwidth provisioning (FBP) achieves fine-grained bandwidth assurance for individual flows. It is especially important for virtualization-based computing environments such as data centers. However, existing flow-level bandwidth provisioning solutions suffer from a number of drawbacks, including high implementation complexity, poor performance guarantees, and inefficiency to process variable length packets. In this paper, we study flow-level bandwidth provisioning for Combined Input Crosspoint Queued (CICQ) switches in the OpenFlow context. First, we propose the Flow-level Bandwidth Provisioning algorithm for CICQ switches, which reduces the switch scheduling problem to multiple instances of fair queuing problems, each utilizing a well-studied fair queuing algorithm. We theoretically prove that FBP can closely emulate the ideal Generalized Processing Sharing model, and accurately guarantee the provisioned bandwidth. Furthermore, we implement FBP in the OpenFlow software switch to obtain realistic performance data by a prototype. Leveraging the capability of OpenFlow to define and manipulate flows, we experimentally demonstrate a practical flow-level bandwidth provisioning solution. Finally, we conduct extensive simulations and experiments to evaluate the design. The simulation data verify the correctness of the analytical results, and show that FBP achieves tight performance guarantees. The experiment results demonstrate that our OpenFlow-based prototype can conveniently and accurately provision bandwidth at the flow level.

We present the design and implementation of FileSim, a simulation framework with detailed models of parallel file systems, capable of reproducing the complex I/O behavior at scale. FileSim aims to support comprehensive and accurate end-to-end I/O performance prediction and evaluation of exascale high-end computing systems. To this end, FileSim provides several key features, including detailed, pluggable models of contemporary parallel file systems, the support of trace-driven simulation, and the capability of running large-scale I/O systems using parallel and distributed simulation.We conducted extensive validation and performance studies, through which we show that the simulator is capable of reproducing important I/O system behaviors comparable to those measured from the real systems. We demonstrate the capabilities of FileSim as a tool for exploring the parameter space and design alternatives of large-scale parallel file systems.

Modern data center networks (DCNs) often use multi-rooted topologies, which offer multipath capability, for increased bandwidth and fault tolerance. However, traditional routing algorithms for the Internet have no or limited support for multipath routing, and cannot fully utilize available bandwidth in such DCNs. In this paper, we study the multipath routing problem for DCNs. We first formulate the problem as an integer linear program, but it is not suitable for fast on-the-fly route calculation. For a practical solution, we propose the Depth-First Worst-Fit Search based multipath routing algorithm. The main idea is to use depth-first search to find a sequence of worst-fit links to connect the source and destination of a flow. Since DCN topologies are usually hierarchical, our algorithm uses depth-first search to quickly traverse between hierarchical layers to find a path. When there are multiple links to a neighboring layer, the worst-fit link selection criterion enables the algorithm to make the selection decision with constant time complexity by leveraging the max-heap data structure, and use a small number of selections to find all the links of a path. Further, worst-fit also achieves load balancing, and thus generates low queueing delay, which is a major component of the end-to-end delay. We have evaluated the proposed algorithm by extensive simulations, and compared its average number of link selections and average end-to-end delay with competing solutions. The simulation results fully demonstrate the superiority of our algorithm and validate the effectiveness of our designs.

Flow level bandwidth provisioning offers fine granularity bandwidth assurance for individual flows. It is especially important for virtual network based experiment environments, to isolate traffic of different experiments or different types, which may be fed to the same switch or router port. Existing flow level bandwidth provisioning solutions suffer from a number of drawbacks, including high implementation complexity, poor performance guarantees, and inefficiency to process variable length packets. In this paper, we study flow level bandwidth provisioning for combined-input-crosspoint-queued switches in the OpenFlow context. We propose the FEBR (Flow lEvel Bandwidth pRovisioning) algorithm, which reduces the switch scheduling problem to multiple instances of fair queueing problems, each employing a well studied fair queueing algorithm. FEBR can tightly emulate the ideal Generalized Processing Sharing model, and accurately guarantee the provisioned bandwidth. Further, we implement FEBR in the OpenFlow version 1.0 software switch. In conjunction with the capability of OpenFlow to flexibly define and manipulate flows, we thus provide a practical flow level bandwidth provisioning solution. Finally, we present extensive simulation and experiment data to validate the analytical results and evaluate our design.

Memory consumption is a critical problem for large-scale network simulations. Particularly, the large memory footprint needed for maintaining routing tables can severely obturate scalability. We present an approach of composing large-scale network models using sharable model fragments to achieve significant reduction in the amount of memory required for storing forwarding tables in simulation. Our approach, called spherical routing, conducts static routing within spheres according to user-defined policies. Our routing scheme pre-calculates the forwarding table for each routing sphere, and allows spheres with identical sub-structures to share forwarding tables. Through extensive experiments we demonstrate that our approach can achieve several orders of magnitude in memory reduction for large-scale network models.

Background traffic has a significant impact on the behavior of network services and protocols. However, a detailed model of the background traffic can be extremely time consuming in simulation. In this paper, we extend our previous hybrid model that combines fluid and packet-oriented characterization of network traffic for a realistic representation of the background traffic on Internet. In particular, we get rid of some unrealistic assumptions in the hybrid model, by adding an acknowledgment scheme to correctly capture the mutual influence of fluid TCP flows on network queues, and by applying the Poisson Pareto Burst Process (PPBP) model to describe the long-range dependencies of the Internet traffic. Experiments show that our fluid background traffic model can capture similar level of realism as the traditional packet-oriented approach.

Immersive real-time large-scale network simulation is a technique that supports simulation of large-scale networks to interact with real implementations of network protocols, network services, and distributed applications. Traffic generated by real network applications is carried by the virtual network simulated in real time where delays and losses are calculated according to the simulated network conditions. We emphasize network immersion so that the virtual network is indistinguishable from a physical testbed in terms of network behavior, yet allows the flexibility of simulation. In this paper we provide a summary of current research in immersive real-time large-scale network simulation, particularly in areas of hybrid network traffic modeling and scalable emulation infrastructure design.

We present a software infrastructure that embeds physical hosts in a simulated network. Aiming to create a large-scale real-time virtual network testbed, our real-time interactive simulation approach combines the advantages of both simulation and emulation, by maintaining flexibility of the simulation models and increasing fidelity as real systems are included in the simulation. In our approach, real-world distributed applications and network services can run together with the real-time simulator; real packets are injected into the simulation and subject to the simulated network conditions computed as a result of both real and virtual traffic competing for network resources. A prototype of the proposed emulation infrastructure has been implemented based on virtual private network (VPN). One distinct advantage of our approach is that it does not require special hardware. Furthermore, it is flexible, secure, and scalable-attributes inherited directly from the VPN implementation. We conducted a set of preliminary experiments to assess the performance limitations of our emulation infrastructure. We also present an interesting case study to demonstrate the capability of our approach.

Most comparisons of wireless ad hoc routing algorithms involve simulated or indoor trial runs, or outdoor runs with only a small number of nodes, potentially leading to an incorrect picture of algorithm performance. In this paper, we report on an outdoor comparison of four different routing algorithms, APRL, AODV, ODMRP, and STARA, running on top of thirty-three 802.11-enabled laptops moving randomly through an athletic field. This comparison provides insight into the behavior of ad hoc routing algorithms at larger real-world scales than have been considered so far. In addition, we compare the outdoor results with both indoor ("tabletop") and simulation results for the same algorithms, examining the differences between the indoor results and the outdoor reality. Finally, we describe the software infrastructure that allowed us to implement the ad hoc routing algorithms in a comparable way, and use the same codebase for indoor, outdoor, and simulated trial runs.

All analytical and simulation research on ad~hoc wireless networks must necessarily model radio propagation using simplifying assumptions. We provide a comprehensive review of six assumptions that are still part of many ad hoc network simulation studies, despite increasing awareness of the need to represent more realistic features, including hills, obstacles, link asymmetries, and unpredictable fading. We use an extensive set of measurements from a large outdoor routing experiment to demonstrate the weakness of these assumptions, and show how these assumptions cause simulation results to differ significantly from experimental results. We close with a series of recommendations for researchers, whether they develop protocols, analytic models, or simulators for ad~hoc wireless networks.

This paper considers a technique for composing global (barrier-style) and local (channel scanning) synchronization protocols within a single parallel discrete-event simulation. Composition is attractive because it allows one to tailor the synchronization mechanism to the model being simulated. We first motivate the problem by showing the large performance gap that can be introduced by a mismatch of model and synchronization method. Our solution calls for each channel between submodels to be classified as synchronous or asynchronous. We mathematically formulate the problem of optimally classifying channels and show that, in principle, the optimal classification can be obtained in time proportional to max{C/spl times/log C, V/spl times/N}, where C is the number of channels, V the number of unique minimal delays on those channels, and N is the number of submodels. We then demonstrate an implementation which finds an optimal solution at runtime and consider its performance on network topologies, including one of the global Internet at the autonomous system level. We find that the automated method effectively determines channel assignments that maximize performance.