Hide Notes 

Slide 1: From Cloud Computing to the New Enterprise Data Center Willy Chiu, Vice President, High Performance On Demand Solutions (HiPODS), IBM Software Group Management contact: Jayashree Subrahmonia , jays@us.ibm Team lead: Dennis Quan, Senior Technical Staff Member HiPODS, dennisq@us.ibm.com. Technical contacts: Catherine C. Diep, cdiep@us.ibm.com Linda Legregni, legregni@us.ibm.com Yuanhui Liu, liuyu@us.ibm.com Rahul Jain, rahulj@us.ibm.com Web address: www.ibm.com/developerworks/websphere/zones/hipods/ Date: 28 May 2008 Status: Version 1.0 Executive sponsor: Abstract: This paper describes how the HiPODS team has harvested workload patterns, best practices, and reusable assets from our worldwide customer cloud centers to build a solutions framework for the new enterprise data center. We describe a high-level infrastructure framework and its underlying technologies. We also present examples of value propositions for new enterprise data centers based on cloud infrastructures that have been built as of today. Copyright IBM Corporation 2008
Slide 2: Executive summary Cloud computing is an emerging model where users can gain access to their applications from anywhere through their connected devices. A simplified user interface makes the infrastructure supporting the applications transparent to users. The applications reside in massively-scalable data centers where compute resources can be dynamically provisioned and shared to achieve significant economies of scale. A strong service management platform results in near-zero incremental management costs when more IT resources are added to the cloud. The proliferation of smart mobile devices, high speed wireless connectivity, and rich browser-based Web 2.0 interfaces has made the network-based cloud computing model not only practical but also a source of reduced IT complexity. Players across the IT industry have announced cloud computing efforts of varying shapes and sizes, leading analysts to attempt to identify various characteristics, such as infrastructure outsourcing, software as a service, and next generation distributed computing, to describe these efforts. Many CIOs are asking IBM how they can adopt cloud computing technologies and management techniques to improve the efficiency and flexibility of their own data centers. IBM, with its pioneering position in the virtualization space dating back to the 1960s, has recently introduced its vision of a new enterprise data center. This vision brings together the strengths of the Web-centric cloud computing model and the enterprise data center of today. It will provide request-driven, dynamic allocation of computing resources for a mix of workloads on a massively scalable, heterogeneous, and virtualized infrastructure. It will also be optimized for security, data integrity, and transaction processing. With experience in both enterprise data centers and cloud computing, IBM is uniquely positioned to provide clients with the best solutions to achieve this vision. The IBM High Performance On Demand Solutions (HiPODS) team has been working with leading edge clients around the world, such as Google and the Government of Wuxi in China, to define best practices for running data centers with workloads ranging from Web 2.0 applications to mission critical transaction processing systems. Specifically, we are working with clients to define and enhance a framework for running large scale data centers that enables key functionality for hosting a wide range of applications. This framework today includes automation for the complex time-consuming processes of provisioning of applications and systems, data center-level virtualization, and support for extremely data-intensive workloads. Customers will want this framework to include reliable, secure transaction processing on service oriented architectures. Principles espoused by the new enterprise data center framework apply equally to handling outsourced or in-house data and application components. This framework was inspired by Web 2.0 customers who often have tens to hundreds of thousands of servers handling billions of hits a day. IBM applied the lessons of these early customers to bring these economies of scale and support for massively scalable computing to a broad range of customers. Our hands-on experience setting up cloud data centers represents a major step in the evolution to a new enterprise data center. This paper describes a high level infrastructure services framework for a new enterprise data center and the underlying technology enablers, such as virtualization, automation, self-service portal, monitoring, and capacity planning. It also discusses examples of and value propositions for the data centers that have been built so far. These data centers can host a mix of workloads from Java™ 2 Enterprise Edition (J2EE) applications to software development, test environments as well as data-intensive business intelligence analytics applications. © IBM Corporation 2008 2 From Cloud Computing to the New Enterprise Data Center
Slide 3: Introduction This section reviews the history of cloud computing and introduces the IBM vision for a new enterprise data center. The next section starting on page 5 introduces an infrastructure framework for a new enterprise data center and discusses the virtualized environment and infrastructure management. The last section starting on page 13 describes existing cloud infrastructures and their application. Evolution of cloud computing Cloud computing is an important topic. However it is not a revolutionary new development. Rather it is an evolution that has taken place over several decades, as shown in Figure 1. Figure 1. Evolution toward cloud computing The trend toward cloud computing started in the late 1980s with the concepts of grid computing when, for the first time, a large number of systems were applied to a single problem, usually scientific. Many people ask us what the difference is between grid and cloud computing. The primary difference is how each provides the needed resources to a workload. • In grid computing, the focus is on the ability of moving a workload to the location of the needed computing resources, which are mostly remote and are readily available for use. Usually a grid is a cluster of servers on which a large task could be divided into smaller © IBM Corporation 2008 3 From Cloud Computing to the New Enterprise Data Center
Slide 4: tasks to run in parallel. From this point of view, a grid could actually be viewed as just one virtual server. Grids also require applications to conform to the grid software interfaces. • In a cloud environment, computing resouces, such as servers, can be dynamically shaped or carved out from its underlying hardware infrastructure and made available to a workload. In addition, while a cloud does support grid, a cloud can also support nongrid environments, such as a three-tier Web architecture running traditional or Web 2.0 applications. In the 1990s the concept of virtualization was expanded beyond virtual servers to higher levels of abstraction, first the virtual platform, and second the virtual application. Utility computing offered clusters as virtual platforms for computing with a metered business model. More recently software as a service (SaaS) raised the level of virtualization to the application, with a business model of charging not by the resources consumed, but by the value of the application to subscribers. The concept of cloud computing has evolved from the concepts of grid, utility, and SaaS. It is an emerging model where users can gain access to their applications from anywhere through their connected devices. These applications reside in massively-scalable data centers where compute resources can be dynamically provisioned and shared to achieve significant economies of scale. The proliferation of smart mobile devices, high speed wireless connectivity, and rich browserbased Web 2.0 interfaces has made the network-based cloud computing model not only practical but also a source of reduced IT complexity. The strength of a cloud is its infrastructure management, enabled by the maturity and progress of virtualization technology to manage and better utilize the underlying resources through automatic provisioning, reimaging, workload rebalancing, monitoring, and systematic change request handling. The new enterpise data center As more and more players across the IT industry announce cloud computing initiatives, many CIOs are asking IBM how they can adopt cloud computing technologies and management techniques to improve the efficiency and flexibility of their own data centers. IBM, with its pioneering position in the virtualization space dating back to the1960s, has recently introduced its vision of a new enterprise data center, which brings together the strengths of both the Webcentric cloud computing model and the enterprise data center of today, as shown in Figure 2. The new enterprise data center will be a virtualized, efficiently-managed center, which will employ some of the tools and techniques adopted by Web-centric clouds, generalized for adoption by a broader range of customers, and enhanced to support secure transactional workloads. With this highly efficient and shared infrastructure, the ability for companies to respond instantaneously to new business needs, to interpret large amounts of information in real time, and to make sound business decisions based on moment-in-time data becomes possible. The new enterprise data center is an evolutionary new model to provide a new scale of efficient and dynamic approach in helping to align IT with business goals. The remaining sections of this paper describe the high level infrastructure services framework for a new enterprise data center and its underlying technology enablers such as virtualization, automation, provisioning, monitoring, and capacity planning.. Finally, it includes examples of actual data center implementations that reveal some insights into which characteristics of new © IBM Corporation 2008 4 From Cloud Computing to the New Enterprise Data Center
Slide 5: enterprise data center models can offer the most value to customers of any size across a variety of usage scenarios. Figure 2. Cloud computing and the new enterprise data center Architecture framework and technology enablers From the high level architectual point of view, new enterprise data center infrastructure services can be logically divided into layers, as shown in Figure 3. The physical hardware layer is virtualized to provide a flexible adaptive platform to improve resource utilization. The keys to new enterprise data center infrastructure services are the next two layers, the virtualization environment and management layer. The combination of these two layers ensure that resources in a data center are efficiently managed and can be provisioned, deployed, and configured rapidly. In addition, a new enterprise data center is designed to handle a mixture of workload patterns, as discussed in the business use cases section of this paper. © IBM Corporation 2008 5 From Cloud Computing to the New Enterprise Data Center
Slide 6: Figure 3. Infrastructure services framework of a new enterprise data center Virtualized environment What is virtualization? Virtualization is the notion that compute resources can be made available with a large degree of agility and flexibility without requiring that users have detailed knowledge of underlying physical resources. In a virtualized environment, computing environments can be dynamically created, expanded, shrunk, or moved as demand varies. Virtualization provides important advantages in sharing, manageability, and isolation (the ability for multiple users and applications to share physical resources without affecting each other) in a cloud environment. Virtualization allows a set of underutilized physical servers to be consolidated into a smaller number of more fully utilized physical servers, contributing to significant cost savings and providing tremendous advantages in a new enterprise data center cloud environment. There are many aspects of virtualization across the IT infrastructure. Virtualization can mean different things to different people depending on the context. A common interpretation of server virtualization is the mapping of a single physical resource to multiple logical representations or partitions. Logical partitions (LPARs) and virtual machines (VMs) are examples of this definition. IBM pioneered this space dating back to the 1960s. Virtualization technology is not limited to servers. Virtualization applies as well to storage, networking and application, which could be the subject of their own papers. © IBM Corporation 2008 6 From Cloud Computing to the New Enterprise Data Center
Slide 7: How does server virtualization work? In most cases, server virtualization is accomplished by the use of a hypervisor to logically assign and separate physical resources. The hypervisor allows the guest operating system, running on the virtual machine, to believe it is running on the actual hardware, unaware that other guests are sharing it. Each guest operating system is protected from the others and is not affected by any instability or configuration issues of the others. Today, hypervisors are becoming a ubiquitous virtualization layer on client and server systems. There are two major types of hypervisors: baremetal and hosted hypervisors. Bare-metal hypervisors A bare-metal hypervisor runs directly on server hardware to provide virtual machines with finegrained timesharing of resources. Examples of firmware-based bare-metal hypervisors include IBM® System z™ Processor Resource System Manager™ (PR/SM™) and the IBM® System i™ and IBM® System p™ POWER™ Hypervisor. Examples of software-based bare-metal hypervisor include z/VM®, VMware ESX Server and Xen Hypervisor. The overhead of firmware-based hypervisors is generally less than the overhead of software-based hypervisors. In addition, virtualization implemented at the server hardware level can provide the highest efficiency and performance. System z PR/SM is a hardware partitioning capability that enables highly scalable, robust, reliable virtual server hosting for multiple operating system images, each in its own LPAR. Virtualization is built in to the System z. System z also provides the software-based z/VM®. Because z/VM virtualization technology enables highly granular sharing of system resources at high levels of efficiency and resource utilization, it allows clients to run hundreds, or even thousands, of Linux servers on a single mainframe running with other System z operating systems. System p enables users to request virtualized LPARs with IBM® AIX® or Linux operating systems. System p has the micropartitioning capability, which allows the system to assign partial CPUs to LPARs. A partial CPU can be as granular as 1/10 of a physical CPU. With the micropartitioning and virtual I/O capability, the number of LPARs supported by a System p is significantly increased. Furthermore, the LPAR’s CPU resource can be managed by IBM® Enterprise Workload Manager™, which monitors CPU demand and usage and employs business policies to determine how much CPU resource is assigned to each LPAR. Micropartitioning and virtual I/O make a powerful virtualized infrastructure available for cloud users. Using technologies such as VMWare and Xen, virtualization is now extended to the x86 platforms. With these technologies, many virtual servers with different operating systems can coexist on and share the resources of a single x86-based server. Consolidation of applications running in heterogeneous operating system environments and on underutilized hardware servers becomes possible. Speedy provisioning is another benefit of using VMWare and Xen. Because a virtual server's operating system images and configuration and log files are stored as file sets on the file system of the host or management console server, operating system images of a virtual machine can be run on one physical server and moved or copied transparently to another. Provisioning of a new server or cloning of an existing server can be accomplished by simply creating a virtual server on an exising physical system and copying the previously saved virtual server images. In other words, a server can be provisioned without reinstalling the operating system or the applications running on it. © IBM Corporation 2008 7 From Cloud Computing to the New Enterprise Data Center
Slide 8: Hosted hypervisors A hosted hypervisor runs on a host operating system and uses operating system services to provide timesharing of resources to virtual machines. Examples of software based hosted hypervisors include VMware Server and Microsoft Virtual Server. Infrastructure management Virtualization is the fundamental technology enabler for the infrastructure services of the new enterprise data center. Equally important is its management layer that acts like the brain or control center to efficiently manage the resources in the whole environment. The following sections review the major components of this layer. Automation Infrastructure administration is one of the major challenges in a virtualization environment. Simply building a virtualization environment without the proper approach to administration can increase complexity, thereby erasing cost savings derived from the benefits of virtualization. Automation is the key to managing these problems. It is critical that a new enterprise data center be equipped with tools that facilitate, simplify, and enable management of the physical environment that provides the virtual server resources. Automated provisioning Automation is an important technique that is applied to two of the most frequent tasks performed in a dynamic data center: the onboarding and offboarding of applications. Onboarding is the process of installing and configuring the operating system and additional software on servers to the point that they can be made available to do useful work. Offboarding refers to the steps necessary to automatically reclaim a server so it’s available for other purposes. Onboarding an application typically starts with provisioning servers. When done manually onboarding is a time and labor consuming process consisting of many complex steps, such as installation of the operating system and software and configuration of the network and storage. These tasks are often error-prone and typically require highly skilled administrators in the area of system, storage, and network. Furthermore, applications typically have unique installation and configuration steps. All these make automation a key for consistently managing these tasks. IBM® Tivoli® Provisioning Manager is a key component of our cloud computing solutions. It enables cloud administrators to write workflows that automate the installation and configuration of new and existing servers. It supports speedy and efficient construction and management of IT resources. Automated reservations and scheduling Critical to administering computing resources is the ability to understand what the current and future capacity is to accommodate new requests. Without that understanding, one cannot accurately forecast how many customers can be supported, nor ensure that a steady pipeline of applications can be maintained. Again, Tivoli Provisioning Manager, which uses IBM® WebSphere® Application Server to communicate the provisioning status and availability of resources, provides the capability to schedule the provisioning and deprovisioning of resources, and to reserve resources for future use. © IBM Corporation 2008 8 From Cloud Computing to the New Enterprise Data Center
Slide 9: Self-service portal A self-service portal provides systematic request handling and change management capabilities. Service providers understand the need to allow customers or customer representatives to request services or view status of currently deployed services through a Web portal. A new enterprise data center must be able to flexibly handle and execute change requests rapidly to align with fast changing business needs. A request driven provisioning system should be employed to take user requests for new services or change requirements for existing services, for example, allowing users to extend the service end date or add or remove resources requirements. Figure 4 shows a sample screenshot for adding resources to a service (project) and Figure 5 shows a sample screenshot for changing the project end date. Figure 4. Sample screenshot for adding resources to a project © IBM Corporation 2008 9 From Cloud Computing to the New Enterprise Data Center
Slide 10: Figure 5. Sample screenshot for changing project end date Monitoring Monitoring resources and application performance is an important element of any environment. The task of monitoring becomes harder, yet more critical, in a virtualized environment. The benefits provided by monitoring include: • • • • • Collecting historic data to assist with planning future data center resource needs and to optimize virtualized resource placement Capturing real-time data to quickly react to unexpected resource needs Measuring adherence to performance service level agreements (SLAs) Proactively generating alerts and detail data to quickly detect and solve application problems Reporting resource usage data by application, necessary for allocating costs appropriately Monitoring can be performed using IBM® Tivoli® Monitoring, which can monitor the health (CPU, disk, and memory) of the servers provisioned by Tivoli Provisioning Manager. This involves installing an IBM Tivoli Monitoring agent on each cloud server and configuring the IBM Tivoli Monitoring server. The agents collect information from cloud resources and periodically transfer that data to the monitoring data warehouse, which is an IBM® DB/2® database. The monitoring server contains three components; IBM® Tivoli® Enterprise Monitoring, IBM® Tivoli® Enterprise Portal, and the data warehouse. With Tivoli Monitoring, either individual servers or collections of servers can be monitored. Detailed information on each monitored resource can be viewed with Tivoli Enterprise portals and could be fully integrated with the cloud portal. Also, summary information denoting server health can be viewed directly from the cloud self-service portal. Figure 6 shows the CPU, memory, and disk summary information that is consolidated at a project level where a project could contain more than one server or resource. © IBM Corporation 2008 10 From Cloud Computing to the New Enterprise Data Center
Slide 11: Figure 6. Projects summary Capacity planning The cloud computing model reduces the need for capacity planning at an application level. An application can request resources from the cloud and obtain them in less than an hour. An application that needs more resources can submit another request and obtain more resources. Thus it is far less important to correctly predict the capacity requirements for an application than it is in traditional data centers, where as many as six months might be needed to order and install hardware dedicated to the application. On the other hand, virtualization makes it harder and more important to plan capacity from the data center’s perspective. In the past the data center could use the projections from the applications, understand the hardware on order, and essentially avoid having to size the hardware itself. Traditionally, a data center would just need to make sure that it had the capability to support the hardware planned by individual applications. But in a cloud environment, many different applications will be installed. It becomes the data center’s responsibility to predict the average or total resource requirement of all the applications and to order enough hardware in advance without relying on the input of application owners. The basis for capacity planning is monitoring existing usage and keeping track over historical time periods. Long term trends can be projected based on previous activity and adjusted with any knowledge of business plans. In a new enterprise data center, typical capacity planning techniques can be applied for the most part. Since clouds use virtualized resources that share the same physical resources, this makes capacity panning somewhat more complex. In contrast, the capacity planning does not need to consider each individual application, but can simply track and project the overall summation of all applications on the cloud. There are still times when an individual application may be significant to the overall data center, or have a need for initial capacity without allowing for its capacity to grow organically over a longer time period. To address this need, IBM uses its Performance and Capacity Estimation Service (PACES), previously known as Sonoma. PACES is a Web-based service for evaluating performance and planning the capacity of business patterns, product workloads, and custom userdefined workloads. The PACES capacity planning engine is based on a mathematical model © IBM Corporation 2008 11 From Cloud Computing to the New Enterprise Data Center
Slide 12: involving queuing theory that has been fine tuned using empirical data from numerous past engagements; IBM regularly validates the underlying model based on the results of recent projects. Currently PACES is used by IBM employees around the world from groups such as IBM TechLine group, IBM Sales, IBM Global Services, and IBM IT architects. Customers come to IBM with an application framework and performance objectives. PACES models these complex relationships and either suggests a satisfying configuration solution, or projects the performance of a user-specified configuration. Today PACES supports a large library of different kinds of workloads along with an extensive library for such hardware as IBM System x™, System p, Sun, and HP. Figure 7 shows a sample PACES objective specification screen, and Figure 8 shows a sample PACES estimate results screen. Figure 7. Sample PACES objective specification screen Figure 8. Sample PACES estimate results screen © IBM Corporation 2008 12 From Cloud Computing to the New Enterprise Data Center
Slide 13: Business use cases With both a long term pioneering position in the virtualization space and decades of experience in designing and running large, mission-critical distributed computing environments for the clients, IBM has recently introduced its vision of a new enterprise data center and has been working closely with clients to implement this model. Within IBM, the High Performance On Demand Solutions (HiPODS) team has been working with leading edge customers around the world, such as Google and the Government of Wuxi in China, to verify best practices for running data centers with Web 2.0 and other types of workloads. The following sections share some of the insights that have been gained. Innovation enablement IBM’s internal innovation portal The office of the IBM CIO created an internal innovation program to enable and foster a culture of innovation and harvest the collective intelligence of our large employee base. This program allows internally-developed projects to be publicized and made available for use by the IBM internal (and self-selected) early adopters, who will provide valuable feedback as they use the new applications. Innovators not only learn what application features and functions users like, they also receive feedback on nonfunctional areas such as performance, scalability, operational guidelines, resource utilization, and support requirements. The innovation program staff collects metrics and analyzes the business value of the program. Previously, a typical pilot team needed four to twelve weeks to identify, procure, and build a pilot infrastructure and additional time to build a security compliant software stack so that developers could begin building or deploying applications. The CIO office needed to dedicate highly-skilled administrators to manage this IT environment. How to reduce IT management costs while accelerating innovation became a key business goal. To achieve this goal, we implemented an internal innovation portal and a virtualized infrastructure dedicated to hosting applications and their associated services environment. Innovators who want to obtain broad, user-based feedback publish information about their application on the portal and deploy their applications in this environment. We provided a rolebased self-service portal that allows innovators to fill out a form defining their hardware platform, CPU, memory, storage, operating system, and middleware, as well as project team members and their associated roles. This process takes about five minutes. After submitting the request through the portal, the program administrator is notified and logs in to approve, modify, and / or reject the request. If approved, the system begins a process to build the server(s). This process involves Web services, Tivoli Provisioning Manager, and optionally IBM Tivoli Security Compliance Manager, business process execution language (BPEL), IBM Enterprise Workload Manager, and Remote Deployment Manager / Cluster Systems Management / Network Installation Manager The process is fully automatic and completes in about an hour or less. The core functionality of this new enterprise data center environment is its ability to automatically provision servers for innovators and to enable them, program administrators, and others to use that function with a Web-based interface. The role-based interface abstracts the capabilities provided by IBM Tivoli Provisioning Manager, Remote Deployment Manager, Network Installation Manager, BPEL, and Web services. © IBM Corporation 2008 13 From Cloud Computing to the New Enterprise Data Center
Slide 14: The value of a fully automated provisioning process that is security compliant and automatically customized to innovators’ needs results in: • Reduced time to introduce technologies and innovations • Cost savings in labor for designing, procuring, and building hardware and software platforms • Cost avoidance through greater use and reuse of existing resources. This business-responsive data center has broken down a critical barrier to enabling innovation. In 2007, the innovation portal hosted 92,823 registered early adopters (almost one third of the IBM employee population). Over seventy different applications were deployed, used, and evaluated. Many of these applications graduated out of the program into production use. Ten of these applications were subsequently turned into IBM products. One of the most successful projects was IBM® Lotus® Sametime® Version 7.5 – Web Conferencing, which had 65,000 registered users providing feedback and comments over blogs and forums offered through the innovation portal. Sogeti innovation cloud Sogeti, a large consulting company in Europe and a subsidiary of CapGemini, wanted their managers to convene at their annual meeting and brainstorm new ideas that the company should implement. But like many other companies, Sogeti was struggling with the question where do we start? They did not have the right tool to enable this kind of dialog among the managers, nor the resources and time to implement a solution on their own. During the progression of ideas through the innovation life cycle, IT resources are often necessary to support the activities. A barrier to this process for Sogeti, as well as many other customers, is the time required to request, instantiate, and support a set of IT systems for an innovation program. Sogeti came to IBM for a solution. By working with different clients to create an innovation program, we have learned that collaboration tools alone do not yield desired results as effectively as a structured innovation platform and program. Figure 9 shows the logical architecture of an innovation platform, including each of the engines required for the application components and features. Information about the application and associated people is stored in a common IBM® Idea Factory database and an LDAP repository, respectively. An optional security authentication and authorization component is shown, as well, for enhanced security and customization for granular content authorization. © IBM Corporation 2008 14 From Cloud Computing to the New Enterprise Data Center
Slide 15: Figure 9. Logical architecture of an innovation platform IBM Idea Factory is a specific solution developed by IBM to implement the generic innovation platform. Idea Factory is an Enterprise Web 2.0 solution that accelerates the innovation life cycle. The innovation life cycle is usually divided into the five phases: ideation, idea selection, development, incubation, and graduation. The goal of Idea Factory is to provide a flexible, dynamic platform for innovators to collaborate on ideas and create customized Web sites for interacting with interested users. For Sogeti, we decided to leverage an IBM Idea Factory solution and host it in a regional IBM cloud data center based on the new enterprise data center model. As mentioned above, one barrier that Sogeti had was that during the progression of ideas through the innovation life cycle, IT resources are often needed to be requested, instantiated, and supported, which requires extensive time and highly skilled IT system administrators. Cloud computing overcomes this barrier perfectly. We decided to host this Idea Factory solution in a cloud environment where VM images representing different components within a standard innovation platform are used to be deployed easily and quickly as needed. By using this virtualization technique within a cloud environment, we were able not only to automate the process and reduce the time and effort needed to provision a new set of J2EE application systems from days or weeks to a matter of minutes or hours, but also to make the whole application easier to be scaled out whenever needed. By virtue of the automation built into the management of the cloud data center, we were able to accommodate this request and provide the Idea Factory solution as a service to Sogeti. By leveraging the IBM Idea Factory solution with a new enterprise data center model, Sogeti achieved its goal to enable its employees to easily share and collaboratively work on the ideas and concepts more effectively. © IBM Corporation 2008 15 From Cloud Computing to the New Enterprise Data Center
Slide 16: Software development and test environments China Cloud Computing Center at Wuxi The city of Wuxi, about 100 miles outside of Shanghai, China, has an economic development project to establish a software park (a specialized zone that provides substantial tax incentives for businesses that open new offices in the software park). One challenge facing startups in Wuxi was the significant high upfront investment in IT infrastructure they needed to make before being able to accept business from enterprise clients. To address this challenge and attract companies to the software park, the municipal government of Wuxi worked with IBM to build a cloud computing center based on a new enterprise data center model. Tenants in the software park can use this data center to rent software development and test environments. Figure 10 shows a logical view for its management and customer environments. Figure 10. Management and customer environments It utilizes Linux kickstart (a scripting tool for automating Linux OS installation), network file system (NFS), logical volume manager (LVM), and either Xen or the p5 hypervisor as the underlying virtualization environment. The management layer consists of Tivoli Provisioning Manager and Tivoli Monitoring. Some easy-to-use Web 2.0 user interface and business management logic components have been implemented for users to easily request and manage projects consisting of collections of virtual machines. © IBM Corporation 2008 16 From Cloud Computing to the New Enterprise Data Center
Slide 17: Besides automatically provisioning operating systems like Linux RedHat Version 5 on both System x and System p, and some middleware like WebSphere Application Server Version 6.1 and DB2 Enterprise Server Edition Version 9.1, this solution also provides the feature to automatically provision some Rational® products like IBM® Rational® for Mulitplatform Version 7.0.1 Multilingual, IBM® Rational® Performance Tester for Mulitplatform Version 7.0 Multilingual, IBM® Rational® PurifyPlus Enterprise Edition for Mulitplatform V 7.0 Multilingual, and IBM® Rational® Software Architect for Mulitplatform Version 7.0.1 Multilingual to provide software companies with the development and test environments they need. Because multiple customers are hosted within one environment, this solution requires strong network isolation and security. In this virtualized environment, hosts from one physical server may have VMs used for multiple projects, and also one project might span multiple hosts. Virtual private network (VPN) technology is used to make sure each client has its own isolated network. When resources are provisioned, additional networks / bridges are configured on either the Xen host or virtual I/O server. Advanced computing model for data-intensive workloads IBM/Google Academic Initiative With more Web 2.0 applications freely available to Internet users and more users uploading audio and video, two challenges are faced by companies like Google and FaceBook: 1) how to reliably store all this data; and 2) how to extract business value from the large volume of daily Web traffic. The MapReduce distributed parallel programming model is increasingly used to write these types of business analytics programs. MapReduce allows the processing to be distributed across hundreds to thousands of nodes, all of them working in parallel on a subset of the data. The intermediate results from these nodes are combined, sorted, and filtered to remove duplicates to arrive at the final answer. The majority of processing at Google is based on this MapReduce model. Google’s success has prompted other companies to follow the same model, and the increasing popularity of MapReduce-style programming inspired the Apache Hadoop project, which is an open-source implementation of the MapReduce programming framework. Many companies use Apache Hadoop for large scale business analytics, ranging from understanding users' navigation patterns and trends on their Web sites to building targeted advertising campaigns. The characteristics of the MapReduce programming model require an underlying compute infrastructure that is highly scalable, and for this reason the new enterprise data center platform provides an ideal foundation for such workloads. To promote this surge of interest in MapReduce style programming, IBM and Google announced a partnership in October 2007 to provide a number of data centers for use by the worldwide academic community. These centers are powered by an IBM solution based on new enterprise data center architecture for data center management, which allows users to quickly provision large Hadoop clusters for students from leading universities, who might be short of required IT resources to otherwise complete their lab assignments or run their research programs. Some of the universities using these centers and teaching courses on the MapReduce methods are University © IBM Corporation 2008 17 From Cloud Computing to the New Enterprise Data Center
Slide 18: of Washington, University of Maryland, Massachusetts Institute of Technology, Carnegie Mellon University, University of California Berkeley, and Colorado State University.. The heart of this solution is to automatically provision a large cluster of virtual machines for students to access through the Internet to test their parallel programming projects. As a result, physical machines, or virtual machines created using the Xen hypervisor, can be provisioned rapidly and automatically using Network Installation Manager, Remote Deployment Manager, or Cluster Systems Manager, depending upon the operating system and platform. The cluster is powered with open source software, including: Linux (Fedora), Xen systems virtualization, and the Hadoop workload scheduler. MapReduce Tools for Eclipse, which is open source software designed by IBM to help students develop programs for clusters running Hadoop, is available in Hadoop 0.16.2. Although the current implementations of this cloud support Xen specifically, the framework also allows for other software virtualization technologies such as VMWare ESX Server. From this joint effort with Google, we have not only gained a much deeper understanding of the characteristics and programming model for data-intensive workloads, for example, business anyalitics applications, but also invaluable hands-on experience with how the new enterprise data center model can be applied to this type of workload. Summary Today’s IT realities make cloud computing a natural for meeting the needs of both: • IT providers who demand unprecedented flexibility and efficiency, lower costs and complexity, and support for varied and huge workloads, and • Internet users who expect availability, function, and speed that knows no bounds As technology such as virtualization and corresponding management services like automation, monitoring, and capacity planning services become more mature, cloud computing will become more widely used for more diverse and “mission-critical” types of workloads. With its distinguished history of managing and supporting the largest and most complex IT infrastructures, and with its long-term pioneering position in virtualization space, IBM is wellpositioned to implement its vision of the new enterprise data center to meet demands and expectations, and to satisfy today’s challenges as well as those of the foreseeable trajectory of needs and expectations. © IBM Corporation 2008 18 From Cloud Computing to the New Enterprise Data Center
Slide 19: References HiPODS papers related to or referred to in this paper include: • • Creating a platform for innovation by leveraging the IBM Idea Factory solution, March 2008 at ibm.com/ibmdl/pub/software/dw/wes/hipods/Idea_Factory_wp_14Mar.pdf Sonoma: Web Service for Estimating Capacity and Performance of Service-Oriented Architecture (SOA) Workloads, October 2006 at www.software.ibm.com/software/dw/wes/hipods/SONOMA_wp9Oct_final.pdf See all the HiPODS white papers at www.ibm.com/developerworks/websphere/zones/hipods/library.html Of particular interest, see also: • IBM’s Vision For The New Enterprise Data Center, March 2008 at www.ibm.com/systems/optimizeit/datacenter/pdf/NEDC_POV_MAR_2008_-_02.pdf Here are links to other related references in this paper: HousingMaps FlickrVision VMWare Vmotion IBM Tivoli Provisioning Manager IBM System Director Active Energy Manager Apache Hadoop MapReduce Green Computing Initiatives IBM-Google academic initiative www.housingmaps.com flickrvision.com vmware.com/products/vi/vc/vmotion.html www.ibm.com/software/tivoli/products/prov-mgr www.ibm.com/systems/management/director/extensions/actengmrg.html hadoop.apache.org MapReduce: Simplified Data Processing on Large Clusters at labs.google.com/papers/mapreduce.html www.ibm.com/technology/greeninnovations/index.html www.ibm.com/jct09002c/university/scholars/skills/internetscale/index.html Acknowledgements We acknowledge this paper’s major supporters, contributors, and reviewers: • • Executive sponsor: Willy Chiu Contributions and feedback from the HiPODS Architecture Review Board: Greg Boss, Jim Busche, Catherine Cuong Diep, Harold Hall, Nick Hill, Chin Huang, Eugene Hung, Rahul Jain, Thomas Pavela, Dennis Quan, Linda Legregni, Yuanhui Liu, John Reif, Animesh Singh, Marcus A. Tylutki, Dennis Quan, Jose Vargas, Noshir Wadia. Other reviewers: Susan Holic. • © IBM Corporation 2008 19 From Cloud Computing to the New Enterprise Data Center
Slide 20: Notices Trademarks The following are trademarks or registered trademarks of International Business Machines Corporation in the United States, other countries, or both: IBM, IBM logo, BladeCenter, DB2, Enterprise Workload Manager, Lotus, POWER5, POWER6, Processor Resource System Manager, PR/SM, Rational, Sametime, System i, System p, System x, System z, Tivoli, WebSphere Java and all Java-based trademarks and logos are trademarks or registered trademarks of Sun Microsystems, Inc. in the United States and other countries. Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft Corporation in the United States, other countries, or both. Linux is a registered trademark of Linus Torvalds in the United States, other countries, or both. Other company, product, and service names may be trademarks or service marks of others. Special Notice The information contained in this document has not been submitted to any formal IBM test and is distributed AS IS. The use of this information or the implementation of any of these techniques is a customer responsibility and depends on the customer’s ability to evaluate and integrate them into the customer’s operational environment. While IBM may have reviewed each item for accuracy in a specific situation, there is no guarantee that the same or similar results will be obtained elsewhere. Anyone attempting to adapt these techniques to their own environments do so at their own risk. While IBM may have reviewed each item for accuracy in a specific situation, IBM offers no guarantee or warranty to any user that the same or similar results will be obtained elsewhere. Any person attempting to adapt the techniques contained in this document to their own environment(s) does so at their own risk. Any performance data contained in this document were determined in various controlled laboratory environments and are for reference purposes only. Customers should not adapt these performance numbers to their own environments as system performance standards. The results that may be obtained in other operating environments may vary significantly. Users of this document should verify the applicable data for their specific environment. © IBM Corporation 2008 20 From Cloud Computing to the New Enterprise Data Center