Performance Evaluation Methodologies and Tools

We consider a user’s communication with a receiver in presence of a jammer, in the most competitive situation for user and jammer when they do not have access to complete information on channel gains of each other although they could have access to exact information on their own channel gains. The problem is modeled as a Bayesian power control game between user and jammer as players. Incomplete information is modeled as statistical data over possible channel gains (also referred as channel states). Since channel gain is a function on the distance to the receiver, this also covers scenarios where the user and jammer could know its own location via global positioning system (GPS), but none of them know the exact location of the other. A novel approach is suggested to derive equilibrium of such problems in closed form for two communication metrics: signal-to-interference-plus-noise ratio (SINR) metric, reflecting regular data transmission, and latency metric, reflecting emergency data transmission. In particular, it is shown that the user’s equilibrium strategies corresponding to the latency metric is more sensitive to the a priori statistical information, as compared to the SINR metric. This reflects an advantage of implementing latency metric in case of availability of exact information on network parameters, and an advantage of implementing SINR metric in case of lack of such its availability.

Blockchains are digital ledgers of transactions that aim to be decentralized, secure, and tamper-proof. To achieve this goal, they rely on a consensus algorithm, with the most well-known being the proof-of-work (PoW) algorithm. In PoW, a group of specialized users known as miners invest a significant amount of energy to secure the blockchain ledger. Miners are incentivized to participate in the network through the potential rewards they can earn, which are based on the number of blocks they are able to consolidate and add to the chain. An important characteristic of the PoW algorithm is that miners’ rewards must be statistically proportional to the amount of computational power (and hence energy) invested in this process. In this work, we study the selfish miner attack by means of a stochastic model based on a quantitative process algebra. When a successful attack occurs, a miner or mining pool is able to receive more rewards than they should, at the expense of other miners. The model analysis allows us to derive the conditions under which the attack becomes convenient for the miners.

Evolution Strategy (ES) is a potent black-box optimization technique based on natural evolution. A key step in each ES iteration is the ranking of candidate solutions based on some fitness score. In the Reinforcement Learning (RL) context, this step entails evaluating several policies. Presently, this evaluation is done via on-policy approaches: each policy’s score is estimated by interacting several times with the environment using that policy. Such ideas lead to wasteful interactions since, once the ranking is done, only the data associated with the top-ranked policies are used for subsequent learning. To improve sample efficiency, we introduce a novel off-policy ranking approach using a local approximation for the fitness function. We demonstrate our idea for two leading ES methods: Augmented Random Search (ARS) and Trust Region Evolution Strategy (TRES). MuJoCo simulations show that, compared to the original methods, our off-policy variants have similar running times for reaching reward thresholds but need only around 70% as much data on average. In fact, in some tasks like HalfCheetah-v3 and Ant-v3, we need just 50% as much data. Notably, our method supports extensive parallelization, enabling our ES variants to be significantly faster than popular non-ES RL methods like TRPO, PPO, and SAC.

We consider a non-cooperative game in the SIR model with confinements. Each member of the population is a player whose strategy is her probability of being protected from the epidemic. We assume that for each player, there is a cost of infection per unit time and a cost of being confined, which is linear and decreasing on her confinement strategy. The total cost is defined as the sum of her confinement and infection costs. We present a method for computing a symmetric Nash equilibrium for this game and study its efficiency. We conclude that the Nash equilibrium we obtain leads to fewer confinements than the strategy that minimizes the cost of the entire population.

Microservices deployed and managed by container orchestration frameworks like Kubernetes are the bases of modern cloud applications. In microservice performance modeling and prediction, simulations provide a lightweight alternative to experimental analysis, which requires dedicated infrastructure and a laborious setup. However, existing simulators cannot run realistic scenarios, as performance-critical orchestration mechanisms (like scheduling or autoscaling) are manually modeled and can consequently not be represented in their full complexity and configuration space. This work combines a state-of-the-art simulation for microservice performance with Kubernetes container orchestration. Hereby, we include the original implementation of Kubernetes artifacts enabling realistic scenarios and testing of orchestration policies with low overhead. In two experiments with Kubernetes’ kube-scheduler and cluster-autoscaler, we demonstrate that our framework can correctly handle different configurations of these orchestration mechanisms boosting both the simulation’s use cases and authenticity.

Controlling computer systems in an optimal way using quantum devices is an important step towards next generation infrastructures that will be able to harness the advantages of quantum computing. While the implications are promising, there is a need for evaluating new such approaches and tools in comparison with prevalent classical alternatives. In this work we contribute in this direction by studying the stabilization and control of a tandem queue system, an exemplary model of a computer system, using model predictive control and quantum annealing. The control inputs are obtained from the minimization of an appropriately constructed cost function and the optimal control problem is converted into a quadratic unconstrained binary optimization problem to be solved by the quantum annealer. We find that as the prediction horizon increases and the core optimization problem becomes complicated, the quantum-enhanced solution is preferable over classical simulated annealing. Moreover, there is a trade-off one should consider in terms of variations in the obtained results, quantum computation times and end-to-end communication times. This work shows a way for further experimentation and exploration of new directions and challenges and underscores the experience gained through utilization of the state-of-the-art quantum devices.

To support reliable and low-latency communication, Time-Sensitive Networking introduced protocols and interfaces for resource allocation in Ethernet. However, the implementation of these allocation algorithms has not yet been covered by the standards. Our work focuses on deadline-guaranteeing resource allocation for networks with static and dynamic traffic. To achieve this, we combine offline network optimization heuristics with online admission control and, thus, allow for new flow registrations while the network is running. We demonstrate our solution on Credit-Based Shaper networks by using the delay analysis framework Network Calculus. We compare our approach with an intuitive and a brute-force algorithm, where we can achieve significant improvements, both, in terms of quality and runtime. Thereby, our results show that we can guarantee maximum end-to-end delays and also increase the flexibility of the network while requiring only minimal user input.

We study content caching in a wireless network in which the users are connected through a base station that is equipped with a finite capacity cache. We assume a fixed set of contents whose popularity vary with time. Users’ requests for the contents depend on their instantaneous popularity levels. Proactively caching contents at the base station incurs a cost but not having requested contents at the base station also incurs a cost. We propose to proactively cache contents at the base station so as to minimize content missing and caching costs. We formulate the problem as a discounted cost Markov decision problem that is a restless multi-armed bandit problem. We provide conditions under which the problem is indexable and also propose a novel approach to manoeuvre a few parameters to render the problem indexable. We demonstrate efficacy of the Whittle index policy via numerical evaluation.

We introduce dSalmon, a highly efficient framework for outlier detection on streaming data. dSalmon can be used with both Python and C++, meeting the requirements of modern data science research. It provides an intuitive interface and has almost no package dependencies. dSalmon implements main stream outlier detection approaches from literature. By using pure C++ in its core and making the most of available parallelism, data is analyzed with superior processing speed.

We describe design decisions and outline the software architecture of dSalmon. Additionally, we perform thorough evaluations on benchmarking datasets to measure execution time, memory requirements and energy consumption when performing outlier detection. Experiments show that dSalmon requires substantially less resources and in most cases is able to process datasets between one and three orders of magnitude faster than established Python implementations.

This paper presents the open-source C++ tool RealySt for effectively computing optimal time-bounded reachability probabilities for subclasses of hybrid automata extended with random clocks. The tool explicitly resolves the underlying nondeterminism and computes reachable state sets exactly. The error of the computed results solely stems from the multi-dimensional integration. The architecture of RealySt is extensible and allows to easily integrate other classes of hybrid automata extended by random clocks. RealySt relies on the HyPro library to perform flowpipe construction, and on GSL for multi-dimensional integration.