Big Data computing and clouds

A goal in cloud computing is to allocate (and thus pay for) only those cloud resources that are truly needed. To date, cloud practitioners have pursued schedule-based (e.g., time-of-day) and rule-based mechanisms to attempt to automate this matching between computing requirements and computing resources. However, most of these "auto-scaling" mechanisms only support simple resource utilization indicators and do not specifically consider both user performance requirements and budget concerns. In this paper, we present an approach whereby the basic computing elements are virtual machines (VMs) of various sizes/costs, jobs are specified as workflows, users specify performance requirements by assigning (soft) deadlines to jobs, and the goal is to ensure all jobs are finished within their deadlines at minimum financial cost. We accomplish our goal by dynamically allocating/deallocating VMs and scheduling tasks on the most cost-efficient instances. We evaluate our approach in four representative cloud workload patterns and show cost savings from 9.8% to 40.4% compared to other approaches.

Auto-scaling to minimize cost and meet application deadlines in cloud workflows

One of many advantages of the cloud is the elasticity, the ability to dynamically acquire or release computing resources in response to demand. However, this elasticity is only meaningful to the cloud users when the acquired Virtual Machines (VMs) can be provisioned in time and be ready to use within the user expectation. The long unexpected VM startup time could result in resource under-provisioning, which will inevitably hurt the application performance. A better understanding of the VM startup time is therefore needed to help cloud users to plan ahead and make in-time resource provisioning decisions. In this paper, we study the startup time of cloud VMs across three real-world cloud providers -- Amazon EC2, Windows Azure and Rackspace. We analyze the relationship between the VM startup time and different factors, such as time of the day, OS image size, instance type, data center location and the number of instances acquired at the same time. We also study the VM startup time of spot instances in EC2, which show a longer waiting time and greater variance compared to on-demand instances.

/pdf/a-performance-study-on-the-vm-startup-time-in-the-cloud-i1mvxpukwk.pdf

A Performance Study on the VM Startup Time in the Cloud

Scalability is said to be one of the major advantages brought by the cloud paradigm and, more specifically, the one that makes it different to an "advanced outsourcing" solution. However, there are some important pending issues before making the dreamed automated scaling for applications come true. In this paper, the most notable initiatives towards whole application scalability in cloud environments are presented. We present relevant efforts at the edge of state of the art technology, providing an encompassing overview of the trends they each follow. We also highlight pending challenges that will likely be addressed in new research efforts and present an ideal scalable cloud system.

/pdf/dynamically-scaling-applications-in-the-cloud-qvimfifrgr.pdf

Dynamically scaling applications in the cloud

Cloud providers possessing large quantities of spare capacity must either incentivize clients to purchase it or suffer losses. Amazon is the first cloud provider to address this challenge, by allowing clients to bid on spare capacity and by granting resources to bidders while their bids exceed a periodically changing spot price. Amazon publicizes the spot price but does not disclose how it is determined.By analyzing the spot price histories of Amazon’s EC2 cloud, we reverse engineer how prices are set and construct a model that generates prices consistent with existing price traces. Our findings suggest that usually prices are not market-driven, as sometimes previously assumed. Rather, they are likely to be generated most of the time at random from within a tight price range via a dynamic hidden reserve price mechanism. Our model could help clients make informed bids, cloud providers design profitable systems, and researchers design pricing algorithms.

/pdf/deconstructing-amazon-ec2-spot-instance-pricing-4sfykd81o9.pdf

Deconstructing Amazon EC2 Spot Instance Pricing

MapReduce is a scalable and fault tolerant framework, patented by Google, for computing embarrassingly parallel reductions. Hadoop is an open-source implementation of Google MapReduce that is made available as a web service to cloud users by the AmazonWeb Services (AWS) cloud computing infrastructure. Amazon Spot Instances (SIs) provide an inexpensive yet transient and market-based option to purchasing virtualized instances for execution in AWS. As opposed to manually controlling when an instance is terminated, SI termination can also occur automatically as a function of the market price and maximum user bid price. We find that we can significantly improve the runtime of MapReduce jobs in our benchmarks by using SIs as accelerators. However, we also find that SI termination due to budget constraints during the job can have adverse affects on the runtime and may cause the user to overpay for their job. We describe new techniques that help reduce such effects.

/pdf/see-spot-run-using-spot-instances-for-mapreduce-workflows-1gakfq3nt4.pdf

See spot run: using spot instances for mapreduce workflows

We present the design and implementation of AppScale, an open source extension to the Google AppEngine (GAE) Platform-as- a-Service (PaaS) cloud technology. Our extensions build upon the GAE SDK to facilitate distributed execution of GAE applications over virtualized cluster resources, including Infrastructure-as-a-Service (IaaS) cloud systems such as Amazon’s AWS/EC2 and Epucalyptus. AppScale provides a framework with which researchers can investigate the interaction between PaaS and IaaS systems as well as the inner workings of, and new technologies for, PaaS cloud technologies using real GAE applications.

/pdf/appscale-scalable-and-open-appengine-application-development-4ib8fpucfl.pdf

AppScale: Scalable and Open AppEngine Application Development and Deployment

We present new cloud support that employs a single API -- the Datastore API from Google App Engine (GAE) -- to interface to different open source distributed database (DB) technologies. We employ this support to ``plug in'' DB technologies to the API so that they can be used by web applications and services without modification. The system facilitates an empirical evaluation and comparison of these disparate systems by web software developers, and reduces the barrier to entry for their use by automating their configuration and deployment.

An Evaluation of Distributed Datastores Using the AppScale Cloud Platform

In this paper, we present the design and implementation of Neptune, a domain specific language (DSL) that automates configuration and deployment of existing HPC software via cloud computing platforms. We integrate Neptune into a popular, open-source cloud platform, and extend the platform with support for user-level and automated placement of cloud services and HPC components. Such platform integration of Neptune facilitates hybrid-cloud application execution as well as portability across disparate cloud fabrics. Supporting additional cloud fabrics through a single interface enables high throughput computing (HTC) to be achieved by users who do not necessarily own grid-level resources but do have access to otherwise independent cloud technologies.We evaluate Neptune using different applications that employ a wide range of popular HPC packages for their implementation including MPI, X10, MapReduce, DFSP, and dwSSA. In addition, we show how Neptune can be extended to support other HPC software and application domains, and thus be used as a mechanism for many task computing (MTC).

/pdf/neptune-a-domain-specific-language-for-deploying-hpc-4yq13jc3sz.pdf

Neptune: a domain specific language for deploying hpc software on cloud platforms

AppScale is a scalable, distributed, and fault tolerant cloud runtime system that we have developed at the University of California, Santa Barbara as part of our research into the next generation of programming systems [5, 3]. In particular, AppScale is a cloud platform, i.e. a platform-as-a-service (PaaS) cloud fabric, that executes over cluster resources. The cluster resources underlying AppScale can be managed with or without virtualization (e.g. Xen, KVM) or via popular cloud infrastructures including Amazon EC2 [6] and Eucalyptus [8]. The AppScale platform virtualizes, abstracts, and multiplexes cloud and system services across multiple applications, enabling write-one, run-anywhere (WORA) program development for the cloud. In addition to simplifying application development and deployment using cloud systems and modern distributed computing technologies, AppScale brings popular public cloud fabrics to “on-premise”, or private, clusters. To enable this, we emulate key cloud layers from the commercial sector – (i) to engender a user community, (ii) to gain access to and to study real applications, and (iii) to investigate the potential implementations of, and extensions to, public cloud systems using open-source technologies. The first API we have chosen to emulate is that of Google App Engine [11]. Google App Engine is a cloud platform that provides scalable implementations of technologies important for web services (messaging, key-value data storage, multi-tasking, web server support, elasticity, and resource management, among others). Using Google App Engine developers debug and test their programs using an open-source software development kit (SDK) provided by Google that implements non-scalable versions of the APIs. Developers then upload their code and data to Google clusters, and can then use Google resources and services “pay-per-use”, on a free or low-cost rental basis. AppScale implements the (open) Google App Engine APIs such that applications that execute on Google App Engine, also execute on AppScale without modification using private cluster resources or public cloud infrastructures. Our API and service implementations target scale, distribution, fault tolerance, high-performance, and high availability. We leverage mature open-source technologies to the greatest degree possible to enable this. AppScale implements multiple language runtimes (Java, Python, Ruby) as application frontends and a wide range of open source database technologies (key-value or relational) as options for its internal system-wide datastore. AppScale is not a replacement for Google App Engine or any other public cloud technology. Instead, AppScale is a robust and extensive distributed system that provisions the resources it is allocated scalably across multiple applications. The AppScale platform exports services and APIs in addition to those provided by Google App Engine. These technologies are important for application domains beyond those of web services, e.g. data analytics and data and computationally intensive applications. AppScale exports these technologies as services, i.e. AppScale “service-izes” libraries, tools, and software packages, including MapReduce, X10, R, and MPI, and, as it does for the Google App Engine APIs, AppScale provides automatic configuration, deployment, and distribution for them, as well as support for their elasticity, load balancing, and fault tolerance. Since AppScale provides a software layer between applications and the implementations of cloud APIs and services, we are also able to employ existing cloud services for the implementations. As such, AppScale provides a hybrid cloud platform – a programming system through which an application can access services from different clouds (a mix of public and private) at different times or concurrently. Such support can be used by developers to move data between clouds, e.g., for disaster recovery, fault tolerance, data backups, to reduce public cloud costs (to use lower-cost alternatives), and to “fall out” from a private cloud with limited resources to a public cloud on-demand.

Navraj Chohan

Papers

See spot run: using spot instances for mapreduce workflows

AppScale: Scalable and Open AppEngine Application Development and Deployment

An Evaluation of Distributed Datastores Using the AppScale Cloud Platform

Neptune: a domain specific language for deploying hpc software on cloud platforms

AppScale: Open-Source Platform-as-a-Service