Skip to main content

Technical paper

Page 1

Fast Data Referenzarchitektur

1

Fast Data Referenzarchitektur

LO GO


2

Fast Data Referenzarchitektur

Chapter 1

Definition This paper provides an overview and guide to SQL Server® Fast Track Data Warehouse, a new set of reference architectures created for scale-up (SMP) SQL Server based data warehouse solutions. It includes a summary of the resources available in the reference configuration, the distinguishing features of the approach, and the steps necessary to take full advantage of the new architectures. The performance and stability of any application solution—whether line of business, transactional, or business intelligence (BI)— hinges on the integration between solution design and hardware platform. Choosing the appropriate solution architecture—especially for BI solutions—requires balancing the application’s intended purpose and expected use with the hardware platform’s components. Poor planning, bad design, and misconfigured or improperly sized hardware often lead to ongoing, unnecessary spending and, even worse, unsuccessful projects. This paper is a companion resource for Microsoft’s new SQL Server Fast Data reference architectures , which provide tested, pre-configured architectures and architectural guidance for a BI solution’s database components and hardware systems. The Fast Track reference configuration resources provide planning tools, architectural decision guidance, best practices for design and tuning, specific hardware configurations from Dell and HP, and the tools needed to create a new reference configuration on any hardware platform. The Fast Track Data Warehouse reference configurations focus on the central database component of a BI solution, commonly called the data warehouse or data mart. These Fast Track reference configurations target solutions

on a single server estimated at up to 32TB of data. The intended audience for this paper includes IT executives and managers, solution architects, IT infrastructure planners, and project managers. If you are responsible for the ownership, management, planning, or design of a BI solution, this paper will give you the skills and resources you need to make the best decisions for your solution. If you’ve faced a performance or scalability problem with an existing application or during the testing phase of a new solution, you’ve probably heard the question, “Why don’t we just buy a bigger server…or get faster hard drives…or add more memory?” Throwing hardware at a problem rarely leads to cost savings or long-term success. Instead of addressing the core problem, reflexively going with “bigger iron” without comprehensive investigation and planning is merely putting a Band-Aid on a deeper issue. So when planning for performance in a BI solution or responding to a performance problem, what is the right answer? Of course, that depends on the type of solution, its intended use, and the actual or anticipated usage load. Although each application has its unique footprint of activity, most applications fall into different general categories of use. Common types of applications have patterns of activity that you can quantify; you can then architect and configure specifically for that type of activity. For example, a BI solution’s activity is very different from that for a transactional solution, such as an ERP system.


Fast Data Referenzarchitektur

3

Chapter 2

Pre-configured A BI system focuses on responding to queries that summarize, filter, and pivot large sets of data. Transactional systems, on the other hand, manage the detail records of activity or events and, therefore, deal with mostly smaller sets of data—typically individual records. The most common mistake in designing a BI solution is configuring the hardware and architecture as if it were a transactional system. The Fast Track reference configurations seek to help you avoid that mistake by providing the architectural guidance and hardware design patterns that enable you to focus on the real purpose of the BI system: adding value to the business.

Figure 1: Fast Data reference architectures focus on the database tier of a BI solution

The rest of this paper outlines the value of the Fast Data reference architectures , their focus, available tools, and how to take full advantage of the reference configurations to build the best BI solution for your organization.

As noted earlier, the Fast Data reference architectures include the hardware configuration, the tools to identify the right system, and best practices for configuring the software and database.

Perhaps the most important benefit of using the Fast Data reference architectures is avoiding the pitfalls of an improperly designed and configured system. Misconfigured hardware will limit the capability and throughput of a system. And users will experience delays or timeouts when trying to gather the analytic output from the system, which will directly impact their adoption of the solution.

Before you can select a sample Fast Data reference architectures or build your own, you first need to identify estimates for the solution, including the expected user count, the complexity of the queries, the estimated data scanned per query type, the concurrency, and the data volume. With these estimates, you can leverage provided calculations to determine the number of CPU cores and the system throughput needed.

The Fast Track reference configurations address the need to design solution architectures correctly for BI systems. More specifically, these Fast Track reference configurations target the database tier of a BI solution, providing the tools and architectural direction customers need when using SQL Server for a data warehouse that will support BI-focused queries.

If you decide to build your own hardware reference configuration, you’ll need to determine the CPU Core Consumption on the hardware that you choose.


4

Fast Data Referenzarchitektur

Chapter 3

Overview The consumption rate is the driving metric for determining the maximum I/O saturation point of the system and is used to identify the correct storage array and how to best configure it. You will also need to determine how much throughput the Storage Array and I/O channel can handle. See the “Building Your Own Hardware Configuration” section for the detailed steps to build your own Fast Track reference architecture. For the hardware architecture, you can use any server hardware and storage system. The reference configurations include all the calculations necessary to model the SMP architectures on any hardware platform. However, as examples and models, Fast Track reference configurations are provided based on two hardware vendors: Dell and HP with other vendors coming soon.

Now, let’s look at a summary of the role of the data warehouse in an overall BI solution and best practices for its design. A data warehouse solution is most often associated with a larger BI strategy. As you plan to leverage the Fast Data Reference Architectures , it’s important to understand the big-picture architecture related to systems that participate in a BI strategy. This section reviews the overall role of the data warehouse in a corporate environment and the database design best practices for the environment.

See the section “Available SQL Server Fast Data Reference Architectures ” for more details about the reference configurations. The final aspect of the configuration involves prescriptive guidance for optimizing the solution through system and server settings and database tuning. This guidance applies to any hardware platform using SQL Server as the database engine for a data warehouse. These best practices relate to multiple system tiers, including the initial setup of files and tables, the extraction, transformation, and loading (ETL) of data, and ongoing maintenance of the system over time to ensure performance. The “Choosing and Implementing the Right Fast Track Architecture” section further clarifies the overall optimization approach and points to the resources available in the Fast Data reference architectures to assist in the process.

Figure 2 above highlighted the typical solution architecture of a BI environment, which includes ETL processes, the temporary storage of data in a staging environment, the data warehouse or data mart, the cubes, and the presentation layer reports, dashboards, and analytic views. For smaller solutions, some of these components can be co-located on the same physical machine. However, a better architecture choice is to distribute the components across separate servers.


Fast Data Referenzarchitektur

5

Chapter 4

Design The term data warehouse is often used generically to refer to a data repository that houses historical business data. Although this is true, it doesn’t capture the design practices that should be used within a data warehouse environment. Data warehouses are frequently designed using a modeling technique called dimensional modeling, which effectively handles the performance and historical goal. Dimensional modeling focuses on optimizing the structures for reporting by creating some redundancy in the text columns (called attributes) and then separating those from the numeric metrics analyzed. Like dimension tables, fact tables also track history, such as the history of sales, inventory quantities, account balances, or events. For example, a store inventory fact table could track the weekly history of products quantities in stock at every store. With this information, users can perform trending and analysis of inventory levels over time—a capability rarely available in a source transactional system. The Fast Track Data Warehouse reference architecture recommendations for indexing and partitioning, outlined in the “Choosing and Implementing the Right Fast Track Architecture” section, focus on tables that are representative of a data warehouse design. The approach Microsoft used for creating its SMP based Fast Track reference configurations for data warehouses can be applied to any hardware platform. “Building Your Own Hardware Configuration,” later in this paper, outlines the steps for creating your own Fast Data reference architectures if you have existing hardware you need to repurpose or your preferred hardware platform isn’t part of this first phase of tested

Fast Track reference architectures. However, if you’re building a new data warehouse or data mart BI solution or have latitude in selecting a new system, the existing Fast Track reference configurations give you the important advantage of extensive testing to balance the hardware components. Let’s explore the distinguishing features of these new Fast Track Data Warehouse architectures, and then review the details of each preconfigured solution so that you can select the right hardware configuration for your solution. Like dimension tables, fact tables also track history, such as the history of sales, inventory quantities, account balances, or events. For example, a store inventory fact table could track the weekly history of products quantities in stock at every store. With this information, users can perform trending and analysis of inventory levels over time—a capability rarely available in a source transactional system. The Fast Track Data Warehouse reference architecture recommendations for indexing and partitioning, outlined in the “Choosing and Implementing the Right Fast Track Architecture” section, focus on tables that are representative of a data warehouse design. The approach Microsoft used for creating its SMP based Fast Track reference configurations for data warehouses can be applied to any hardware platform. “Building Your Own Hardware Configuration,” later in this paper, outlines the steps for creating your own Fast Data reference architectures if you have existing hardware you need to repurpose


6

Fast Data Referenzarchitektur


Turn static files into dynamic content formats.

Create a flipbook
Technical paper by JPKusmin - Issuu