Björn Frits Postema

An AnyLogic Simulation Model for Power and Performance Analysis of Data Centres

In this paper we propose a simulation framework that allows for the analysis of power and performance trade-offs for data centres that save energy via power management. The models are cooperating discrete-event and agent-based models, which enable a variety of data centre configurations, including various infrastructural choices, workload models, (heterogeneous) servers and power management strategies. The capabilities of our modelling and simulation approach is shown with an example of a 200-server cluster. A validation that compares our results, for a restricted model with a previously published numerical model is also provided.

Fluid Survival Tool: A Model Checker for Hybrid Petri Nets

Recently, algorithms for model checking Stochastic Time Logic (STL) on Hybrid Petri nets with a single general one-shot transition (HPNG) have been introduced. This paper presents a tool for model checking HPNG models against STL formulas. A graphical user interface (GUI) not only helps to demonstrate and validate existing algorithms, it also eases use. From the output of the model checker, 2D and 3D plots can be generated. The extendable object-oriented tool has been developed using the Model-View-Controller and Facade patterns, Doxygen for documentation and Qt for GUI development written in C++.

Specification of Data Centre Power Management Strategies

In recent work, we proposed a flexible simulation framework (using AnyLogic) for the trade-off analysis of power and performance in data centres. We now extend this framework with a versatile module to study the effect of advanced power management strategies based on both power and performance measurement data collected during system operation. This allows us to study the effect of power management in more detail by taking a wide variety of state information variables into account. The paper includes examples of several power management strategies presented in the literature (and extensions thereof). By enabling power management, our more advanced strategies show energy consumption reductions of up to 55% for a typical data centre workload for a small 30 servers cluster, while performance is kept intact and Service-Level Agreements violations are minimised.

Stochastic Petri Net Models for the Analysis of Trade-Offs in Data Centres with Power Management

Due to the growth in energy consumption of data centres, the demand for optimal usage of servers has become a relevant topic. This paper contributes to the early design phases of data centres by providing insight into the power-performance trade-off that arises from power management. This paper proposes a flexible set of stochastic Petri net models which can be used easily to study the trade-off between performance and power consumption.

Evaluating Load Balancing Policies for Performance and Energy-Efficiency

Nowadays, more and more increasingly hard computations are performed in challenging fields like weather forecasting, oil and gas exploration, and cryptanalysis. Many of such computations can be implemented using a computer cluster with a large number of servers. Incoming computation requests are then, via a so-called load balancing policy, distributed over the servers to ensure optimal performance. Additionally, being able to switch-off some servers during low period of workload, gives potential to reduced energy consumption. Therefore, load balancing forms, albeit indirectly, a trade-off between performance and energy consumption. In this paper, we introduce a syntax for load-balancing policies to dynamically select a server for each request based on relevant criteria, including the number of jobs queued in servers, power states of servers, and transition delays between power states of servers. To evaluate many policies, we implement two load balancers in: (i) iDSL, a language and tool-chain for evaluating service-oriented systems, and (ii) a simulation framework in AnyLogic. Both implementations are successfully validated by comparison of the results.

Combining Energy Saving Techniques in Data Centres using Model-Based Analysis

Advanced power management and cooling techniques for data centres often co-exist as separate entities in current-day operation of data centres. This paper proposes to combine these techniques to achieve greater power savings. To this end, an existing theoretical thermal-aware model is integrated in an extensive simulation framework for data centres using power and performance models, which allows for a detailed study in power, performance and thermal metrics. The paper compares four distinct cases for studying the effect on these metrics: a data centre with (i) basic functionality; (ii) advanced cooling; (iii) advanced power management; and (iv) a combination thereof. The combined case shows a significant reduction in the energy consumption compared to the other cases while performance and thermal demands are kept intact. The combination of these techniques shows improvements in energy savings and shows it is meaningful to investigate further into smart combined energy saving techniques.

Fitting Realistic Data Centre Workloads: A Data Science Approach

Data centres are playing a pivotal role in all cloud-based services (e-commerce, social networks, financial services, e-government, etc.). The performance of data centres is crucial for the acceptance of all these services by end-users. It is important to carefully design data centres with both performance and energy considerations in mind, as data centres are also known to use large amounts of electrical energy. For that purpose we have developed a modular simulation model (based on Anylogic) that can be used to study performance-energy trade-offs in data centre design. Key to such studies is the availability of a workload model. In this paper we present a workload characterisation model and algorithm using modern-day data science techniques, building on top of Jupyter Notebook and the ProFiDo platform. We present the method and show its versatility on a case study with real-world traces of 20 million entries, provided by the Dutch company better.be.

Capabilities of Raspberry Pi 2 for Big Data and Video Streaming Applications in Data Centres

Many new data centres have been built in recent years in order to keep up with the rising demand for server capacity. These data centres require a lot of electrical energy and cooling. Big data and video streaming are two heavily used applications in data centres. This paper experimentally investigates the possibilities and benefits of using cheap, low power and widely supported hardware in the form of a micro data centre with big data and video streaming as its main application area. For this purpose, multiple Raspberry Pi 2 Model B (RPi2)’s have been used in order to build a fully functional distributed Hadoop and video streaming setup that has acceptable performance and extends to new research opportunities. We experimentally validated the new setup to fit in a data centre environment by analysis of its performance, scalability, energy consumption, temperature and manageability. This paper proposes a high concurrency and low power setup in a small 1U form factor with an estimated number of 72 RPi2’s as an interesting alternative to traditional rack servers.