Cloud Workloads Research Articles

Dynamic Heterogeneity-Aware Resource Provisioning in the Cloud

Data centers consume tremendous amounts of energy in terms of power distribution and cooling. Dynamic capacity provisioning is a promising approach for reducing energy consumption by dynamically adjusting the number of active machines to match resource demands. However, despite extensive studies of the problem, existing solutions have not fully considered the heterogeneity of both workload and machine hardware found in production environments. In particular, production data centers often comprise heterogeneous machines with different capacities and energy consumption characteristics. Meanwhile, the production cloud workloads typically consist of diverse applications with different priorities, performance and resource requirements. Failure to consider the heterogeneity of both machines and workloads will lead to both sub-optimal energy-savings and long scheduling delays, due to incompatibility between workload requirements and the resources offered by the provisioned machines. To address this limitation, we present Harmony, a Heterogeneity-Aware dynamic capacity provisioning scheme for cloud data centers. Specifically, we first use the K-means clustering algorithm to divide workload into distinct task classes with similar characteristics in terms of resource and performance requirements. Then we present a technique that dynamically adjusting the number of machines to minimize total energy consumption and scheduling delay. Simulations using traces from a Google's compute cluster demonstrate Harmony can reduce energy by 28 percent compared to heterogeneity-oblivious solutions.

QoS-Aware scheduling in heterogeneous datacenters with paragon

Large-scale datacenters (DCs) host tens of thousands of diverse applications each day. However, interference between colocated workloads and the difficulty of matching applications to one of the many hardware platforms available can degrade performance, violating the quality of service (QoS) guarantees that many cloud workloads require. While previous work has identified the impact of heterogeneity and interference, existing solutions are computationally intensive, cannot be applied online, and do not scale beyond a few applications. We present Paragon, an online and scalable DC scheduler that is heterogeneity- and interference-aware. Paragon is derived from robust analytical methods, and instead of profiling each application in detail, it leverages information the system already has about applications it has previously seen. It uses collaborative filtering techniques to quickly and accurately classify an unknown incoming workload with respect to heterogeneity and interference in multiple shared resources. It does so by identifying similarities to previously scheduled applications. The classification allows Paragon to greedily schedule applications in a manner that minimizes interference and maximizes server utilization. After the initial application placement, Paragon monitors application behavior and adjusts the scheduling decisions at runtime to avoid performance degradations. Additionally, we design ARQ, a multiclass admission control protocol that constrains application waiting time. ARQ queues applications in separate classes based on the type of resources they need and avoids long queueing delays for easy-to-satisfy workloads in highly-loaded scenarios. Paragon scales to tens of thousands of servers and applications with marginal scheduling overheads in terms of time or state. We evaluate Paragon with a wide range of workload scenarios, on both small and large-scale systems, including 1,000 servers on EC2. For a 2,500-workload scenario, Paragon enforces performance guarantees for 91% of applications, while significantly improving utilization. In comparison, heterogeneity-oblivious, interference-oblivious, and least-loaded schedulers only provide similar guarantees for 14%, 11%, and 3% of workloads. The differences are more striking in oversubscribed scenarios where resource efficiency is more critical.

Cloud Workload Characterization

This paper describes various computer system workloads and relates them to their underlying resource utilization. Specifically, the paper concentrates on Cloud workload characterization based upon issues, capabilities, and technologies surrounding the categories from the multiple points of view of the various players involved in Cloud Computing. The relationship is established between the categories and key limiting underlying technologies, and the dynamic and measurable low-level metrics and measurements that are used to detect and reduce resource contention, and identify category changes during run-time. Research questions are posed on dynamic low-level measurements and a usage case example with high performance computing (HPC) clusters. The Cloud workload categories can provide a basis for common communication for various viewpoints from players, including facility managers, Cloud IT or service providers, Cloud users, consumers, IT managers, and hardware vendors. This common communication tool will facilitate better service-level agreements (SLAs), capital purchase decisions, and future computer architecture design-decisions.

Hint-Based Execution of Workloads in Clouds with Nefeli

Infrastructure-as-a-Service clouds offer entire virtual infrastructures for distributed processing while concealing all physical underlying machinery. Current cloud interface abstractions restrict users from providing information regarding usage patterns of their requested virtual machines (VMs). In this paper, we propose Nefeli, a virtual infrastructure gateway that lifts this restriction. Through Nefeli, cloud consumers provide deployment hints on the possible mapping of VMs to physical nodes. Such hints include the collocation and anticollocation of VMs, the existence of potential performance bottlenecks, the presence of underlying hardware features (e.g., high availability), the proximity of certain VMs to data repositories, or any other information that would contribute in a more effective placement of VMs to physical hosting nodes. Consumers designate only properties of their virtual infrastructure and remain at all times agnostic to the cloud internal physical characteristics. The set of consumer-provided hints is augmented with high-level placement policies specified by the cloud administration. Placement policies and hints form a constraint satisfaction problem that when solved, yields the final VM-to-host placement. As workloads executed by the cloud may change over time, VM-to-host mappings must follow suit. To this end, Nefeli captures such events, changes VM deployment, helps avoid bottlenecks, and ultimately, improves the quality of the rendered services. Using our prototype, we examine overheads involved and show significant improvements in terms of time needed to execute scientific and real application workloads. We also demonstrate how power-aware policies may reduce the energy consumption of the physical installation. Finally, we compare Nefeli's placement choices with those attained by the open-source cloud middleware, OpenNebula.

Paragon

Large-scale datacenters (DCs) host tens of thousands of diverse applications each day. However, interference between colocated workloads and the difficulty to match applications to one of the many hardware platforms available can degrade performance, violating the quality of service (QoS) guarantees that many cloud workloads require. While previous work has identified the impact of heterogeneity and interference, existing solutions are computationally intensive, cannot be applied online and do not scale beyond few applications. We present Paragon, an online and scalable DC scheduler that is heterogeneity and interference-aware. Paragon is derived from robust analytical methods and instead of profiling each application in detail, it leverages information the system already has about applications it has previously seen. It uses collaborative filtering techniques to quickly and accurately classify an unknown, incoming workload with respect to heterogeneity and interference in multiple shared resources, by identifying similarities to previously scheduled applications. The classification allows Paragon to greedily schedule applications in a manner that minimizes interference and maximizes server utilization. Paragon scales to tens of thousands of servers with marginal scheduling overheads in terms of time or state. We evaluate Paragon with a wide range of workload scenarios, on both small and large-scale systems, including 1,000 servers on EC2. For a 2,500-workload scenario, Paragon enforces performance guarantees for 91% of applications, while significantly improving utilization. In comparison, heterogeneity-oblivious, interference-oblivious and least-loaded schedulers only provide similar guarantees for 14%, 11% and 3% of workloads. The differences are more striking in oversubscribed scenarios where resource efficiency is more critical.

Paragon

Large-scale datacenters (DCs) host tens of thousands of diverse applications each day. However, interference between colocated workloads and the difficulty to match applications to one of the many hardware platforms available can degrade performance, violating the quality of service (QoS) guarantees that many cloud workloads require. While previous work has identified the impact of heterogeneity and interference, existing solutions are computationally intensive, cannot be applied online and do not scale beyond few applications. We present Paragon, an online and scalable DC scheduler that is heterogeneity and interference-aware. Paragon is derived from robust analytical methods and instead of profiling each application in detail, it leverages information the system already has about applications it has previously seen. It uses collaborative filtering techniques to quickly and accurately classify an unknown, incoming workload with respect to heterogeneity and interference in multiple shared resources, by identifying similarities to previously scheduled applications. The classification allows Paragon to greedily schedule applications in a manner that minimizes interference and maximizes server utilization. Paragon scales to tens of thousands of servers with marginal scheduling overheads in terms of time or state. We evaluate Paragon with a wide range of workload scenarios, on both small and large-scale systems, including 1,000 servers on EC2. For a 2,500-workload scenario, Paragon enforces performance guarantees for 91% of applications, while significantly improving utilization. In comparison, heterogeneity-oblivious, interference-oblivious and least-loaded schedulers only provide similar guarantees for 14%, 11% and 3% of workloads. The differences are more striking in oversubscribed scenarios where resource efficiency is more critical.

Hybrid Attribute- and Re-Encryption-Based Key Management for Secure and Scalable Mobile Applications in Clouds

Outsourcing data to the cloud are beneficial for reasons of economy, scalability, and accessibility, but significant technical challenges remain. Sensitive data stored in the cloud must be protected from being read in the clear by a cloud provider that is honest-but-curious. Additionally, cloud-based data are increasingly being accessed by resource-constrained mobile devices for which the processing and communication cost must be minimized. Novel modifications to attribute-based encryption are proposed to allow authorized users access to cloud data based on the satisfaction of required attributes such that the higher computational load from cryptographic operations is assigned to the cloud provider and the total communication cost is lowered for the mobile user. Furthermore, data re-encryption may be optionally performed by the cloud provider to reduce the expense of user revocation in a mobile user environment while preserving the privacy of user data stored in the cloud. The proposed protocol has been realized on commercially popular mobile and cloud platforms to demonstrate real-world benchmarks that show the efficacy of the scheme. A simulation calibrated with the benchmark results shows the scalability potential of the scheme in the context of a realistic workload in a mobile cloud computing system.

Analyzing performance and power efficiency of network processing over 10 GbE

Ethernet continues to be the most widely used network architecture today for its low cost and backward compatibility with the existing Ethernet infrastructure. Driven by increasing networking demands of cloud workloads, network speed rapidly migrates from 1 to 10 Gbps and beyond. Ethernet’s ubiquity and its continuously increasing rate motivate us to fully understand high speed network processing performance and its power efficiency.In this paper, we begin with per-packet processing overhead breakdown on Intel Xeon servers with 10 GbE networking. We find that besides data copy, the driver and buffer release, unexpectedly take 46% of the processing time for large I/O sizes and even 54% for small I/O sizes. To further understand the overheads, we manually instrument the 10 GbE NIC driver and OS kernel along the packet processing path using hardware performance counters (PMU). Our fine-grained instrumentation pinpoints the performance bottlenecks, which were never reported before. In addition to detailed performance analysis, we also examine power consumption of network processing over 10 GbE by using a power analyzer. Then, we use an external Data Acquisition System (DAQ) to obtain a breakdown of power consumption for individual hardware components such as CPU, memory and NIC, and obtain several interesting observations. Our detailed performance and power analysis guides us to design a more processing- and power-efficient server I/O architecture for high speed networks.

Hadoop and memcached: Performance and power characterization and analysis

Given the rapid expansion in cloud computing in the past few years, there is a driving necessity of having cloud workloads running on a backend servers analyzed and characterized for performance and power consumption. In this research, we focus on Hadoop framework and Memcached, which are distributed model frameworks for processing large scale data intensive applications for different purposes. Hadoop is used for short jobs requiring low response time; it is a popular open source implementation of MapReduce for the analysis of large datasets, while Memcached is a high performance distributed memory object caching system that could speed up throughput of web applications by reducing the effect of bottlenecks on database load. In this paper, we characterize different workloads running on Hadoop framework and Memcached for different processor configurations and microarchitecture parameters. We implement an analytical estimation model for performance and power using different server processor microarchitecture parameters. The proposed analytical estimation model uses analytical method to scale different processor microarchitecture parameters such as CPI with respect to processor core frequency. We also propose an analytical model to estimate power consumption scaling for different processor core frequency. The combination of both performance and power consumption analytical models enables the estimation of performance per watt for different cloud benchmarks. The proposed estimation models are verified to estimate power and performance with less than 10% error deviation.

IPaaS: Different Ways of Thinking

The concept of iPaaS is in its infancy, and very few vendors and users even recognize the term. Some see iPaaS as an evolution of integration as a service, which has been widely adopted for cloud services integration. Another point of view presents iPaaS as an emerging form of delivering application and data integration capabilities, consolidating multiple cloud services in a suite aimed at the integration and governance of any combination of on-premises and off-premises applications, within or across organizations. The iPaaS is designed to act in the middle and help connect to a number of services across the layers. Issues come from the convergence and consolidation process toward iPaaS, which will be turbulent. Many providers will disappear; in the race to leadership, vendors will bring to market immature technologies; some mergers and acquisitions will fail; some vendors will not be able to effectively scale up their platforms to support large cloud workloads; and other vendors will struggle to provide quality support to a growing number of clients.

On Cloud computational models and the heterogeneity challenge

AbstractCloud computing is by far the most cost-effective technology for hosting Internet-scale services and applications. The MapReduce model, in particular, is largely used nowadays in Cloud infrastructures to meet the demand of large-scale data and computation intensive applications. Despite its success, the implications of MapReduce on the management of Cloud workload and cluster resources are still largely unstudied. In this article, we show that dealing with the heterogeneity of workloads and machine capabilities is a key challenge. In today’s cloud environment, workloads can have varied sizes, lengths, resource requirements, and arrival rates. The machines also have varied CPU, memory, I/O speed, and network bandwidth capacities. Jointly they pose difficult challenges pertaining, among others, to job scheduling, task and data placement, resource sharing and resource allocation. We analyze the heterogeneity challenge in these specific problem domains and survey the representative state-of-the-art works that try to address them. We found that although advances are made that partially address some of the outlined challenges, there are even more open challenges yet to be explored, and this topic at large is ripe for scientific contributions.

Managing dynamic enterprise and urgent workloads on clouds using layered queuing and historical performance models

The automatic allocation of enterprise workload to resources can be enhanced by being able to make what–if response time predictions whilst different allocations are being considered. We experimentally investigate an historical and a layered queuing performance model and show how they can provide a good level of support for a dynamic-urgent cloud environment. Using this we define, implement and experimentally investigate the effectiveness of a prediction-based cloud workload and resource management algorithm. Based on these experimental analyses we: (i) comparatively evaluate the layered queuing and historical techniques; (ii) evaluate the effectiveness of the management algorithm in different operating scenarios; and (iii) provide guidance on using prediction-based workload and resource management.

Synthetic Workload Generation for Cloud Computing Applications

We present techniques for characterization, modeling and generation of workloads for cloud computing applications. Methods for capturing the workloads of cloud computing applications in two different models - benchmark application and workload models are described. We give the design and implementation of a synthetic workload generator that accepts the benchmark and workload model specifications generated by the characterization and modeling of workloads of cloud computing applications. We propose the Georgia Tech Cloud Workload Specification Language (GT-CWSL) that provides a structured way for specification of application workloads. The GT-CWSL combines the specifications of benchmark and workload models to create workload specifications that are used by a synthetic workload generator to generate synthetic workloads for performance evaluation of cloud computing applications.

Data Management in the Cloud: Promises, State-of-the-art, and Open Questions

Cloud Computing has the potential to significantly change the IT world. It promises dramatic reductions in cost and time-to-market. This paper gives an introduction to cloud computing, thereby stating how cloud computing tries to fulfill these promises. Furthermore, this paper studies the state-of-the-art of cloud computing platforms and answers the question of how well the current generation of systems meets these promises. Based on that analysis, this paper states several fundamental questions that need to be addressed when designing a cloud computing platform. The focus of the paper is on database workloads; more specifically, on online transaction processing workloads (OLTP) in public clouds.