Cloud datacenters provide a backbone to our digital society. Inaccurate capacity procurement for cloud datacenters can lead to significant performance degradation, denser targets for failure, and unsustainable energy consumption. Although this activity is core to improving cloud infrastructure, relatively few comprehensive approaches and support tools exist for mid-tier operators, leaving many planners with merely rule-of-thumb judgement. We derive requirements from a unique survey of experts in charge of diverse datacenters in several countries. We propose Capelin, a data-driven, scenario-based capacity planning system for mid-tier cloud datacenters. Capelin introduces the notion of portfolios of scenarios, which it leverages in its probing for alternative capacity-plans. At the core of the system, a trace-based, discrete-event simulator enables the exploration of different possible topologies, with support for scaling the volume, variety, and velocity of resources, and for horizontal (scale-out) and vertical (scale-up) scaling. Capelin compares alternative topologies and for each gives detailed quantitative operational information, which could facilitate human decisions of capacity planning. We implement and open-source Capelin, and show through comprehensive trace-based experiments it can aid practitioners. The results give evidence that reasonable choices can be worse by a factor of 1.5-2.0 than the best, in terms of performance degradation or energy consumption.

Graphs are ubiquitous abstractions enabling reusable computing tools for graph processing with applications in every domain. Diverse workloads, standard models and languages, algebraic frameworks, and suitable and reproducible performance metrics will be at the core of graph processing ecosystems in the future. Academics, start-ups, and big tech companies such as Google, Face book, and Microsoft have introduced various systems for managing and processing the growing presence of big graphs. An increasing number of use cases revealed RDBMS performance problems in managing highly connected data, motivating various startups and innovative products, such as Neo4j, Sparksee, and the current Amazon Neptune. Microsoft Trinity along with Azure SQL DB have provided an early distributed database-oriented approach to big graph management.

High-quality designs of distributed systems and services are essential for our digital economy and society. Threatening to slow down the stream of working designs, we identify the mounting pressure of scale and complexity of (eco-)systems, of ill-defined and wicked problems, and of unclear processes, methods, and tools. We envision design itself as a core research topic in distributed systems, to understand and improve the science and practice of distributed (eco-)system design. Toward this vision, we propose the AtLarge design framework, accompanied by a set of 8 core design principles. We also propose 10 key challenges, which we hope the community can address in the following 5 years. In our experience so far, the proposed framework and principles are practical, and lead to pragmatic and innovative designs for large-scale distributed systems.

Big data processing systems are becoming increasingly more present in cloud workloads. Consequently, they are starting to incorporate more sophisticated mechanisms from traditional database and distributed systems. We focus in this work on the use of caching policies, which for big data raise important new challenges. Not only they must respond to new variants of the trade-off between hit rate, response time, and the space consumed by the cache, but they must do so at possibly higher volume and velocity than web and database workloads. Previous caching policies have not been tested experimentally with big data workloads. We address these challenges in this work. We propose the Read Density family of policies, which is a principled approach to quantify the utility of cached objects through a family of utility functions that depend on the frequency of reads of an object. We further design the Approximate Histogram, which is a policy-based technique based on an array of counters. This technique promises to achieve runtime-space efficient computation of the metric required by the cache policy. We evaluate through trace-based simulation the caching policies from the Read Density family, and compare them with over ten state-of-the-art alternatives. We use two workload traces representative for big data processing, collected from commercial Spark and MapReduce deployments. While we achieve comparable performance to the state-of-art with less parameters, meaningful performance improvement for big data workloads remain elusive.

Datacenters act as cloud-infrastructure to stakeholders across industry, government, and academia. To meet growing demand yet operate efficiently, datacenter operators employ increasingly more sophisticated scheduling systems, mechanisms, and policies. Although many scheduling techniques already exist, relatively little research has gone into the abstraction of the scheduling process itself, hampering design, tuning, and comparison of existing techniques. In this work, we propose a reference architecture for datacenter schedulers. The architecture follows five design principles: components with clearly distinct responsibilities, grouping of related components where possible, separation of mechanism from policy, scheduling as complex workflow, and hierarchical multi-scheduler structure. To demonstrate the validity of the reference architecture, we map to it state-of-the-art datacenter schedulers. We find scheduler-stages are commonly underspecified in peer-reviewed publications. Through trace-based simulation and real-world experiments, we show underspecification of scheduler-stages can lead to significant variations in performance.

Resource contention is one of the major problems in cloud datacenters. Many types of resource contention occur, with important impact on the performance and sometimes even the reliability of applications running in cloud datacenters. Cloud applications run together on the same physical machines with different workloads resulting in non-synchronized accesses to the shared resources. This leads to cases where co-hosted applications are contending for the common resources and not receiving the demanded resource amounts. In this work, we investigate the contention in CPU resources, as CPU is allowed to be over-committed by typical SLAs. We propose a CPU-contention predictor for the demanding business-critical workloads, which require low resource contention to deliver the required performance to customers. Our predictor is based on a set of regression models and metrics which we evaluate extensively. We tune the predictor with data collected from a real-world cloud operation spanning multiple datacenters and servicing business-critical workloads.

The proliferation of big data processing platforms has led to radically different system designs, such as MapReduce and the newer Spark. Understanding the workloads of such systems facilitates tuning and could foster new designs. However, whereas MapReduce workloads have been characterized extensively, relatively little public knowledge exists about the characteristics of Spark workloads in representative environments. To address this problem, in this work we collect and analyze a 6-month Spark workload from a major provider of big data processing services, Databricks. Our analysis focuses on a number of key features, such as the long-term trends of reads and modifications, the statistical properties of reads, and the popularity of clusters and of file formats. Overall, we present numerous findings that could form the basis of new systems studies and designs. Our quantitative evidence and its analysis suggest the existence of daily and weekly load imbalances, of heavy-tailed and bursty behaviour, of the relative rarity of modifications, and of proliferation of big data specific formats.

Online gaming applications entertain hundreds of millions of daily active players and often feature vastly complex architecture. Among online games, Minecraft-like games simulate unique (e.g., modifiable) environments, are virally popular, and are increasingly provided as a service. However, the performance of Minecraft-like services, and in particular their scalability, is not well understood. Moreover, currently no benchmark exists for Minecraft-like games. Addressing this knowledge gap, in this work we design and use the Yardstick benchmark to analyze the performance of Minecraft-like services. Yardstick is based on an operational model that captures salient characteristics of Minecraft-like services. As input workload, Yardstick captures important features, such as the most-popular maps used within the Minecraft community. Yardstick captures system- and application-level metrics, and derives from them service-level metrics such as frequency of game-updates under scalable workload. We implement Yardstick, and, through real-world experiments in our clusters, we explore the performance and scalability of popular Minecraft-like servers, including the official vanilla server, and the community-developed servers Spigot and Glowstone. Our findings indicate the scalability limits of these servers, that Minecraft-like services are poorly parallelized, and that Glowstone is the least viable option among those tested.

Mobile gaming is already a popular and lucrative market. However, the low performance and reduced power capacity of mobile devices severely limit the complexity of mobile games and the duration of their game sessions. To mitigate these issues, in this article, we explore using computation‐offloading, that is, allowing the compute‐intensive parts of mobile games to execute on remote infrastructure. Computation‐offloading raises the combined challenge of addressing the trade‐offs between performance and power‐consumption while also keeping the game playable. We propose Mirror, a system for computation‐offloading that supports the demanding performance requirements of sophisticated mobile games. Mirror proposes several conceptual contributions: support for fine‐grained partitioning, both offline (set by developers) and dynamic (policy‐based), and real‐time asynchronous offloading and user‐input synchronization protocols that enable Mirror‐based systems to bound the delays introduced by offloading and thus to achieve adequate performance. Mirror is compatible with all games that are tick‐based and user‐input deterministic. We implement a real‐world prototype of Mirror and apply it to the real‐world, complex, popular game OpenTTD. The experimental results show that, in comparison with the non‐offloaded OpenTTD, Mirror‐ed OpenTTD can significantly improve performance and power consumption while also delivering smooth gameplay. As a trade‐off, Mirror introduces acceptable delay on user inputs.

Elasticity is one of the main features of cloud computing allowing customers to scale their resources based on the workload. Many autoscalers have been proposed in the past decade to decide on behalf of cloud customers when and how to provision resources to a cloud application based on the workload utilizing cloud elasticity features. However, in prior work, when a new policy is proposed, it is seldom compared to the state-of-the-art, and is often compared only to static provisioning using a predefined quality of service target. This reduces the ability of cloud customers and of cloud operators to choose and deploy an autoscaling policy, as there is seldom enough analysis on the performance of the autoscalers in different operating conditions and with different applications. In our work, we conduct an experimental performance evaluation of autoscaling policies, using as application model workflows, a popular formalism for automating resource management for applications with well-defined yet complex structures. We present a detailed comparative study of general state-of-the-art autoscaling policies, along with two new workflow-specific policies. To understand the performance differences between the seven policies, we conduct various experiments and compare their performance in both pairwise and group comparisons. We report both individual and aggregated metrics. As many workflows have deadline requirements on the tasks, we study the effect of autoscaling on workflow deadlines. Additionally, we look into the effect of autoscaling on the accounted and hourly based charged costs, and we evaluate performance variability caused by the autoscaler selection for each group of workflow sizes. Our results highlight the trade-offs between the suggested policies, how they can impact meeting the deadlines, and how they perform in different operating conditions, thus enabling a better understanding of the current state-of-the-art.

In the late-1950s, leasing time on an IBM 704 cost hundreds of dollars per minute. Today, cloud computing, that is, using IT as a service, on-demand and pay-per-use, is a widely used computing paradigm that offers large economies of scale. Born from a need to make platform as a service (PaaS) more accessible, fine-grained, and affordable, serverless computing has garnered interest from both industry and academia. This article aims to give an understanding of these early days of serverless computing: what it is, where it comes from, what is the current status of serverless technology, and what are its main obstacles and opportunities.

Graphs are a natural fit for modeling concepts used in solving diverse problems in science, commerce, engineering, and governance. Responding to the variety of graph data and algorithms, many parallel and distributed graph processing systems exist. However, until now these platforms use a static model of deployment: they only run on a pre-defined set of machines. This raises many conceptual and pragmatic issues, including misfit with the highly dynamic nature of graph processing, and could lead to resource waste and high operational costs. In contrast, in this work we explore a dynamic model of deployment. We first characterize workload dynamicity, beyond mere active-vertex variability. Then, to conduct an in-depth elasticity study of distributed graph processing, we build a prototype, JoyGraph, which is the first such system that implements complex, policy-based, and fine-grained elasticity. Using the state-of-the-art LDBC Graphalytics benchmark and the SPEC Cloud Group's elasticity metrics, we show the benefits of elasticity in graph processing: (i) improved resource utilization, (ii) reduced operational costs, and (iii) aligned operation-workload dynamicity. Furthermore, we explore the cost of elasticity in graph processing. We identify a key drawback: although elasticity does not degrade application throughput, graph-processing workloads are sensitive to data movement while leasing or releasing resources.

Within the vast and rich field of online gaming, a new generation of Online Social Games (OSGs) is emerging that have in common a core of social interaction, sometimes explicit, other times implicit. This common core of social experience promises to become at least as important as the experience derived from the game-­‐world itself. In this chapter, we consider the social side of OSGs and provide the following contributions: 1.We motivate the importance of taking social features into account to improve the quality of experience in online gaming. 2.We discuss the various dimensions of (player experience in) OSGs. 3.We describe a social network analysis methodology for identifying relations in OSGs and indicate how this methodology could be used to improve the game-­‐play experience. 4.We also consider and illustrate how certain “social” behaviour, like toxicity, is negative and may harm the game-­‐play experience, if not adequately addressed. 5.We mention several directions for future research to put the power of social features in OSGs to good use.

Graphs are a natural fit for modeling concepts used in solving diverse problems in science, commerce, engineering, and governance. Responding to the diversity of graph data and algorithms, many parallel and distributed graph-processing systems exist. However, until now these platforms use a static model of deployment: they only run on a pre-defined set of machines. This raises many conceptual and pragmatic issues, including misfit with the highly dynamic nature of graph processing, and could lead to resource waste and high operational costs. In contrast, in this work we explore the benefits and drawbacks of the dynamic model of deployment. Building a threelayer benchmarking framework for assessing elasticity in graph analytics, we conduct an in-depth elasticity study of distributed graph processing. Our framework is composed of state-ofthe-art workloads, autoscalers, and metrics, derived from the LDBC Graphalytics benchmark and SPEC RG Cloud Group’s elasticity metrics. We uncover the benefits and cost of elasticity in graph processing: while elasticity allows for fine-grained resource management, and does not degrade application performance, we find that graph workloads are sensitive to data migration while leasing or releasing resources. Moreover, we identify non-trivial interactions between scaling policies and graph workloads, which add an extra level of complexity to resource management and scheduling for graph processing.

Our society is digital: industry, science, governance, and individuals depend, often transparently, on the inter-operation of large numbers of distributed computer systems. Although the society takes them almost for granted, these computer ecosystems are not available for all, may not be affordable for long, and raise numerous other research challenges. Inspired by these challenges and by our experience with distributed computer systems, we envision Massivizing Computer Systems, a domain of computer science focusing on understanding, controlling, and evolving successfully such ecosystems. Beyond establishing and growing a body of knowledge about computer ecosystems and their constituent systems, the community in this domain should also aim to educate many about design and engineering for this domain, and all people about its principles. This is a call to the entire community: there is much to discover and achieve.

The question 'Can big data and HPC infrastructure converge?' has important implications for many operators and clients of modern computing. However, answering it is challenging. The hardware is currently different, and fast evolving: big data uses machines with modest numbers of fat cores per socket, large caches, and much memory, whereas HPC uses machines with larger numbers of (thinner) cores, non-trivial NUMA architectures, and fast interconnects. In this work, we investigate the convergence of big data and HPC infrastructure for one of the most challenging application domains, the highly irregular graph processing. We contrast through a systematic, experimental study of over 300,000 core-hours the performance of a modern multicore, Intel Knights Landing (KNL) and of traditional big data hardware, in processing representative graph workloads using state-of-the-art graph analytics platforms. The experimental results indicate KNL is convergence-ready, performance-wise, but only after extensive and expert-level tuning of software and hardware parameters.

Complex workflows that process sensor data are useful for industrial infrastructure management and diagnosis. Although running such workflows in clouds promises reduces operational costs, there are still numerous scheduling challenges to overcome. Such complex workflows are dynamic, exhibit periodic patterns, and combine diverse task groupings and requirements. In this work, we propose ANANKE, a scheduling system addressing these challenges. Our approach extends the state-of-the-art in portfolio scheduling for datacenters with a reinforcement-learning technique, and proposes various scheduling policies for managing complex workflows. Portfolio scheduling addresses the dynamic aspect of the workload. Reinforcement learning, based in this work on Q-learning, allows our approach to adapt to the periodic patterns of the workload, and to tune the other configuration parameters. The proposed policies are heuristics that guide the provisioning process, and map workflow tasks to the provisioned cloud resources. Through real-world experiments based on real and synthetic industrial workloads, we analyze and compare our prototype implementation of ANANKE with a system without portfolio scheduling (baseline) and with a system equipped with a standard portfolio scheduler. Overall, our experimental results give evidence that a learningbased portfolio scheduler can perform better (5–20%) and cost less (20–35%) than the considered alternatives.

In recent years, many distributed graph-processing systems have been designed and developed to analyze large-scale graphs. For all distributed graph-processing systems, partitioning graphs is a key part of processing and an important aspect to achieve good processing performance. To keep low the overhead of partitioning graphs, even when processing the ever-increasing modern graphs ,many previous studies use lightweight streaming graph-partitioning policies. Although many such policies exist, currently there is no comprehensive study of their impact on load balancing and communication overheads, and on the overall performance of graph-processing systems. This relative lack of understanding hampers the development and tuning of new streaming policies, and could limit the entire research community to the existing classes of policies. We address these issues in this work. We begin by modeling the execution time of distributed graph-processing systems. By analyzing this model under the load of realistic graph-data characteristics, we propose a method to identify important performance issues and then design new streaming graph-partitioning policies to address them. By using three typical large-scale graphs and three popular graph-processing algorithms, we conduct comprehensive experiments to study the performance of our and of many alternative streaming policies on a real distributed graph-processing system. We also explore the impact on performance of using different real-world networks and of other real-world technical details. We further discuss how to use our results, the coverage of our model and method, and
the design of future partitioning policies.

The low performance and power limitations of mobile devices severely limit the complexity and the duration of playing sessions of mobile games. This article examines the possibility of using computation-offloading to mitigate these problems while keeping the game playable. We design Mirror, a framework for offloading computation targeted at the demanding performance requirements of sophisticated mobile games. The key conceptual contributions of Mirror are design decisions that allow for dynamic fine-grained client-side offloading decisions, and a protocol for real-time asynchronous offloading for bounding network delays. We implement a prototype of Mirror and test it by performing offloading for the game OpenTTD. The results are promising, showing that Mirror can increase the performance and decrease the power consumption of games while keeping the gameplay fairly smooth.

In the new Digital Economy, massive computer systems, often grouped in datacenters, serve as factories "producing" cloud services with massive consumption. However, to afford cloud services globally, we must address new research challenges in designing, operating, and using modern datacenters. We must also address challenges in educating and training the next generation of datacenter engineers. Addressing such challenges, in this work we present our vision on OpenDC: we envision the exploration of various datacenter concepts and technologies, using existing and new scientific methods, enabling new education practices and topics, and leading to the creation of new software and data artifacts. We present the datacenter concepts and technologies we are currently planning to explore using OpenDC. We identify the scientific methods we want to use, and explain our vision of education practices. We present the architecture and open-source program underlying the OpenDC software, and the format and open-access data we use for datacenter experiments. We conclude with an open invitation for the community to join our effort.