Context: High computing system performance depends on the interaction between software and hardware layers in modern computer systems. Two strong trends that effect different layers in computer systems are that single processors are now more or less completely replaced by multiprocessors, which are often organized into clusters, and virtualization of resources. The performance evaluation of different software on such physical and virtualized resources, is the focus of this thesis.

Objectives: The objectives of this thesis are to investigate the performance evaluation of SQL and No SQL database management systems, namely Cassandra, CouchDB, MongoDB, PostgreSQL, and RethinkDB; and soft real-time application namely, voice-driven web. Scheduling algorithms for resource allocation for hard real-time applications on virtual processor are also investigated.

Methods: Experiment is used to measure the performance of SQL and No SQL management systems on cluster. It is also used to develop a prototype and predicts processor performance of voice-driven web on multiprocessors. Theoretical methods are used to model and design algorithms to schedule real-time applications on the virtual processor machine. Simulation is used to quantify the performance implications of certain parameter values in our theoretical results and to compare expected performance with theoretical bounds in our schedulability tests.

Results:The performance of Cassandra, CouchDB, MongoDB, 2

PostgreSQL, and RethinkDB is evaluated in terms of writing and reading throughput and latencies in cluster computing. For reading throughput, all database systems are horizontally scalable as the cluster’s nodes number increases, however, only Cassandra and couchDB exhibit scalability for data writing. The overall evaluation shows that Cassandra has the most writing scalable throughput as the number of nodes increases with a relative low latency, whereas PostgreSQL has the lowest writing latency, and MongoDB has the lowest reading latency.

The architectures’ tradeoffs of voice-driven web show that the voice engine should be installed on the server instead of being on the mobile device, and performance evaluations show that speech engine scales with respect to the number of cores in the multiprocessor with and without hyperthreading.

The thesis presents scheduling techniques for real-time applications that runs in virtual machines which are time sharing the processor. Each virtual machine’s period and execution time that allow real-time applications to meet their deadlines can be defined using these techniques. Simulation results show the impact of the length of different VM periods with respect to overhead. The tradeoffs between resources consumption and period length are also given. Furthermore, a utilization based test for scheduling real-time application on virtual multiprocessor is presented. This test determines if a task set is schedulable or not. If the task set is schedulable the algorithm provides the priority for each task. This algorithm avoids Dhall’s effect, which may cause task sets with even very low utilization to miss deadlines.

Conclusions: The thesis presented the performance evaluation of reading and writing throughput and latencies for SQL and NoSQL management systems in the cluster computing. The thesis quantifies the tradeoffs of voice-driven web architectures and the performance scalability of the speech engine with respect to number of cores of the multiprocessor. Furthermore, this thesis proposes scheduling algorithms for real-time 3

application with hard deadline on virtual processors, either as a single core processor or as a multicore processor.

We consider a real-time application that executes in a VM with multiple virtual cores. Tasks are scheduled globally using fixed-priority scheduling. In order to avoid Dhalls effect, we classify tasks into two priority classes: heavy and light. Heavy tasks have higher priority than light tasks. For light tasks we use rate monotonic priority assignment. We propose a utilization-based schedulability test. If the task set is schedulable, we provide an assignment of priorities to tasks. The input to the test is the task set, the number of cores in the VM, the period, deadline and blocking time for the VM. We evaluate how jitter, when scheduling VMs on the hypervisor level, affects the schedulability of the real-time tasks running in the VM. The schedulability of the real-time tasks in the VM decreases when the hypervisor jitter increases, but on the other hand the schedulability on the hypervisor level increases if we allow more jitter, i.e. there is a trade-off. Our results make it possible to evaluate this trade-off and take informed decisions when selecting scheduling parameters on the hypervisor level. Simulations show that the priority assignment used by our algorithm schedules more task sets than using rate monotonic priority assignment. © 2019 The British Computer Society. All rights reserved.

This paper quantifies the performance of the core part of voice driven web using free and open source speech engine; the speech engine which is very high computation demanding, it consists of Automatic Speech Recognition (ASR) and Text To Speech (TTS). Two open source programs, Sphinx-4 and FreeTTS-1.2.2 are used for ASR and TTS respectively. These two programs are executed on 2 different hardware multicore processors with 4 hyperthreaded cores, and 8 cores respectively. The response time with respect to the load variance and the number of cores is measured and predicted using a linear regression model. The results show that, the response time is linear with respect to the input length, this property can be used to directly predict the response for any input length. Moreover, though the response time and the speed up increases as the number of cores increases, the regression coefficients and number of threads reveal that ASR benefits from multicore. The speedup factor for ASR is 1.56 for 8 cores. However for FreeTTS, though being sequential the speed up from the program itself is insignificant, there is about 1. 43 speedup for 8 cores, that comes from the system's contribution. Our findings show that the generalization of the results for multicore processor does not apply to hyperthreading. This paper presents the investigation that is useful for educators, researchers, and applications' developer in voice based applications 'domain.

This paper quantifies the implications and trade-offs of three different architectures for voice driven web application, architectures are implemented as prototypes. The prototypes differ from each other by either using recording, or Text To Speech (TTS) as server based, or TTS as client based to process output speech. A typical application used in this paper, is the most dynamic weather information source which is presented as web feeds or Really Simple Syndication (RSS) feeds. The evaluated quality attributes are performance, maintainability, and development effort. The empirical results show that, each system's architecture has a different quality profile, for instance, one architecture has the lowest development time but the highest maintainability cost, and another has the lowest bandwidth requirements but the highest development cost. Finally, suggestions about optimal choice of system architecture according to the quality requirements of the final system are drawn.