Track finding in particle physics has been an increasing challenge over the past decades because the intensities of collisions in state-of-the-art particle colliders have increased enormously. The particle physics experiments ATLAS and CMS at the CERN LHC have been recording data since 2010. The analysis of the data includes finding the particle tracks. For this, dedicated track reconstruction algorithms were developed and applied. One of the track reconstruction steps is the track finding. For this, track seeds are determined and the tracks are successively reconstructed by adding more measurements. The track finding becomes more complex with an increasing number of concurrently existing tracks created by particles that need to be reconstructed. In the past decade, the intensities of the proton-proton collisions at the LHC increased by more than a factor of ten. The time needed for the track finding increased exponentially due to its combinatorial complexity. The next major challenge for track finding will be the HLLHC operation, with an anticipated 200 concurrent collisions every 25 nanoseconds compared to the current 60 collisions. In this paper, we explain the main ideas of the state-of-the-art GPU-accelerated implementation of the track finding algorithm and describe several optimisations focusing on an early elimination of fake tracks. We compared the time needed to find all concurrently existing tracks of the optimised version with the baseline algorithm. The optimisations reduced the wall time by up to 15 % and the GPU memory requirements by up to 65 % for the anticipated collision rate.

Quantifying the predictive uncertainty of a model is essential for risk assessment.We address the proper calibration of the predictive uncertainty in regression tasks by employing the probability integral transform (PIT) histogram to diagnose miscalibration.PIT histograms are often difficult to interpret, and therefore we present an approach to an automatic interpretation of PIT histograms based on an interpreter trained with a synthetic data set.Given a PIT histogram of a model and a data set, the interpreter can estimate the data-generating distribution of the data set with the main purpose of identifying the cause of miscalibration.* Ondřej Podsztavek, Alexander I. Jordan, and Kai L. Polsterer gratefully acknowledge the generous and invaluable

A key responsibility for Data center (DC) operators is monitoring, which involves analyzing logs, aggregating traffic statistics, and assessing hardware utilization. Monitoring is vital for troubleshooting and predicting traffic patterns. This paper introduces a new highly scalable, easy configurable framework for utilizing sFlow technology to collect and process network statistics of DC network communication, like connection durations, detection of traffic anomalies, etc. We tested the proposed framework in production, thanks to a partnership with a privately owned DC that provides Infrastructure as a Service worldwide.

This paper proposes a novel approach to web service fuzzing that utilizes the OpenAPI Specification. The proposed smart black-box generation-based fuzzer, named openapi-fuzzer, generates and minimizes random payloads to detect vulnerabilities in web services. It is able to minimize the bug-triggering payload to its canonical form. Due to this minimization, it is easy to detect the root cause of an underlying bug. To evaluate its performance, openapi-fuzzer was tested on 3 relevant web services. Kubernetes, Hashicorp Vault, and Gitea. The results demonstrate that openapi-fuzzer outperforms other state-of-the-art web service fuzzers in terms of the number of bugs found and running time.Furthermore, openapi-fuzzer conducts a performance analysis to identify endpoints that are susceptible to Denial-of-Service attacks. By providing developers with detailed statistics, openapi-fuzzer helps them identify and fix performance issues in their web services.

Computations with tensors are widespread in many scientific areas. Usually, the used tensors are very large but sparse, i.e., the vast majority of their elements are zero. The space complexity of sparse tensor storage formats varies significantly. For overall efficiency, it is important to reduce the execution time and additional space requirements of the initial preprocessing (i.e., converting tensors from common storage formats to the given internal format).

In this paper, we demonstrate that neural networks (NNs) are capable of trust estimation and evaluation in ad-hoc networks. The concept of trust in distributed systems arose from the notion of social trust. By the trust problem, we understand the problem of measuring the confidence in the fact that individual nodes behave correctly. We model trust in ad-hoc networks using the packet delivery ratio (PDR) metric. We have developed a method to apply NNs for solving the trust problem in ad-hoc networks. We have conducted a series of simulation experiments and measured the quality of our new method. The results show in average 98% accuracy of the classification and 94% of the regression problem. An important contribution of our research is a verification of the hypothesis that synthetic generation of ad-hoc network traffic in a simulator is sufficient for training of a NN that is then capable to accurately estimate trust in an ad-hoc network.

As contemporary distributed systems reach their scalability limits, their architects search ways to push their boundaries further. One of the approaches is to convert a given distributed system into a peer-to-peer (P2P) structure. This approach causes more overhead compared to a single-master or multi-master architecture, but on larger scales (if properly designed) does not reach any boundaries. The Clondike project started as a grid computing system, but now it aims to create a universal non-dedicated P2P cluster where every participating node can benefit from its membership in the cluster. The P2P approach eliminates the single-point-of-failure problems and high availability is guaranteed by design. So far in the Clondike architecture, a bootstrap protocol was based on broadcasts, which became a limiting factor for the scalability of the whole cluster. We have compared existing P2P communication protocols used in a variety of P2P systems and contemporary P2P file-sharing protocols in order to find a suitable solution for Clondike. We have chosen Kademlia protocol and proven its logaritmic scalability in the Clondike environment. This paper presents results of our analysis and implementation and a series of measurement results comparing the old and new communication protocols.

When authors present new storage formats for sparse matrices, they usually focus mainly on a single evaluation criterion, which is the performance of sparse matrix-vector multiplication (SpMV) in FLOPS. Though such an evaluation is essential, it does not allow to directly compare the presented format with its competitors. Moreover, in case that matrices are within an HPC application constructed in different formats, this criterion alone is not sufficient for the key decision whether or not to convert them into the presented format for the SpMV-based application phase. We establish ten evaluation criteria for sparse matrix storage formats, discuss their advantages and disadvantages, and provide general suggestions for format authors/evaluators to make their work more valuable for the HPC community.

An algorithm for parallel generation of a random permutation of a large set of distinct integers is presented. This algorithm is designed for massively parallel systems with distributed memory architectures and the MPI-based runtime environments. Scalability of the algorithm is analyzed according to the memory and communication requirements. An implementation of the algorithm in a form of a software library based on the C++ programming language and the MPI application programming interface is further provided. Finally, performed experiments are described and their results discussed. The biggest of these experiments resulted in a generation of a random permutation of 2 41 integers in slightly more than four minutes using 131072 CPU cores.

We live in a world where is no problem to own quite powerful workstation. However, despite of such concepts like volunteer computing, grid and cluster computing, there is still a high number of idle workstations that are wasting their computational potentials. We believe that better utilization of an idle computing power can be achieved by economical stimulations. In our recent work, we designed a platform for general purpose computing based on a concept of non-dedicated cluster. In this paper, we proposes a market-based batch system concept, where any cluster user can register his task for an execution on a task board. Any processor can than retrieve the task, execute it and write back its result for a price it chooses. This price can be negotiated by better offer from other processor in a cluster. We implemented this mechanism and made a couple of experiments, that show this concept as a quite promising to be used in a system of resource distribution for non-dedicated cluster.

Techniques for building local area network clusters differ from those used for large scale harvesting of idle computing power. The overlapping non dedicated clusters (ONDC) architecture is trying to mix advantages of both techniques. Clusters build in ONDC style can be deployed both on small scale local networks, but as well in large scale over the Internet deployments. In this paper we analyze the security requirements of ONDC and describe our solution. The solution was implemented for the Clondike clustering system, but the same approach can be used for any other ONDC system. The proposed system is inspired by the security mechanisms of existing P2P grid systems and various trust management systems, but it is tailored to exactly match the ONDC requirements. An important aspect of the proposed solution is a combination of locally issued identity based access control and certificates based delegated authorizations. While the identity based access control is the most intuitive way how to express trust for the users of the system, delegated authorizations are an answer for more complex cases. The unique feature of the proposed system is configurable mechanism for authorization distributions and storage that enables users to trade-off a local information availability with local storage requirements.

Clusters built from commodity computers are popular because of their cost efficiency. There are two basic types of such clusters. A prevalent type is a dedicated cluster where all machines participating in the cluster belong to a single administrative domain. Then there are non-dedicated clusters, where the cluster itself forms its own administrative domain, but the participating computers retain their own administrative domains. Non-dedicated clusters have a better potential for attracting computer owners to participate in a cluster, as they do not have to give up control over their computers. On the other hand, these clusters have much higher security demands than the dedicated ones. In this paper, we focus on the security issues of distributed file systems in the environment of the non-dedicated clusters. We will identify the issues implied by the environment, propose our solution, and demonstrate the feasibility by the experimental implementation.

Clusters of workstations are a promising platform for high-performance computing. Most of such clusters are built of dedicated workstations that cannot be used for other purposes. Even though these clusters offer good price/performance ratio, their costs are not negligible. On the other hand, there exist many idle workstations connected via computer networks. The idea of exploiting such idle resources is obvious. In this paper, we describe an architecture of clusters made of non-dedicated idle workstations. These clusters aim at providing a single-system-image Linux environment. The main innovative feature is that a cluster administration is separate from administration of individual workstations. We discuss issues of security, availability, performance, and administration of such clusters. We also describe a pilot implementation and give promising experimental results.

PC clusters typically operate under continuously changing conditions and load balancing is critically important for efficient utilization of their resources, for maximizing their performance, and for minimizing the process response times. The average process response time is usually considered the most important value for measuring the actual performance of a multitasking system. Clearly, load balancing is critical for clusters running long-term processes (run times of hours). The main motivation of this paper is to study the suitability of the load balancing of short-term CPU-intensive processes, e.g., compilers, compression utilities, etc. Such processes represent a large part of a typical workload of a Linux workstation and load balancing can improve the process response time considerably in such cases. This topic appears to be neglected in the literature. The load balancing can be implemented by process migration (when a running process is stopped, checkpointed, checkpoint data are transferred to another node, and the process is restarted there) or by remote execution (when a process is transferred to another node when it is started and has no allocated memory). While the former solution is more universal (the process can be transferred in any moment), the latter one causes less overhead, which is very important especially when short processes are considered. A load-balancing algorithm is composed of three basic parts: a transferring mechanism, which implements remote execution or process migration, a load-balancing strategy, which makes decisions which process is to be moved where, and an information policy, which collects the information about the cluster state and distributes it among all the nodes. In Linux, a new process is created by fork system call, which makes an exact copy of the parent process, includ-

The mesh-of-trees topology has very attractive properties if a distance-sensitive routing, such as store-and-forward, is used. How its properties compare to meshes if distance-insensitive routing, such as wormhole, is used, was not previously understood. In this paper, we show that meshes of trees allow an elegant and optimal one-to-all broadcast algorithm supposing that routers implement distance-insensitive routing and have all-port capability.

The main results of this paper are algorithms for time-optimal gossip of large packets in noncombining full-duplex all-port 2-D tori and meshes of any size m/spl times/n. The gossip algorithms define the structure of broadcast trees and lock-step scheduling schemes for packets that make the broadcast trees time-are-disjoint. The gossip algorithm for tori is also buffer-optimal-it requires routers with auxiliary buffers for at most three packets. The gossip algorithm for meshes requires routers with auxiliary buffers for O(m+n) packets.

This paper deals with problems of decomposition of de Bruijn graph into isomorphic building blocks based on cover sets. The aim is to find so called lowest-cost cover sets which provide decompositions into building blocks with minimal number of external edges. We present formulae for the costs of basic covers. We give new results on the lowest-cost cover set design. We also give new results on the topology of graphs of building blocks based on basic cover sets. We also discuss several open problems.