Performance Measurements for BI Systems
There are two key measurements for BI systems: the speed at which data is prepared for user queries, and the number of users that can query the system concurrently with acceptable response times.
The nature of BI requires that data must be moved from operational systems to BI systems (data warehouses, data marts, or decision support systems). The data is moved not just so the company has a separate copy to query against; through ETL operations, the data is copied into a database with a star or snowflake schema for both usability and performance reasons. ETL operations can involve large amounts of data, and often companies try to perform these tasks during off-hours to minimize the impact on other operations of the business. This presents a common operational question: can the daily (or weekly, and so on) updates be performed during the time available? To address this question, we will measure the update rate for the system. We measure the update rate by the number of fact table rows updated per hour, because fact tables are larger than dimension tables by orders of magnitude.
Active User Sessions
We build BI systems so that users can query them. There are many ways to think about query performance. OLTP benchmarks such as the TPC-C study the rate at which large volumes of simple queries are processed, measuring the number of queries per second. But these benchmarks do not address the needs of BI systems—BI queries are more complex and address much more data and they are read-only. In addition, the typical BI system supports a smaller set of users than an OLTP system, although deployments to thousands of users are increasing.
Although BI queries are larger and more complex, they must run in a short amount of time (preferably in less than a second) in order to encourage users to explore the data. Response time is a critical measure of the performance of a BI system under load. Many IT organizations have a service level agreement with users, stating that the average response time must be less than a certain threshold, such as five seconds.
The BI Accelerator is a new approach for boosting SAP NetWeaver BI performance based on SAP's search and classification engine, and on specially configured hardware. SAP NetWeaver BI customers adopting the BI Accelerator can expect radical improvements in query performance through sophisticated in-memory data compression and horizontal and vertical data partitioning, with near zero administrative overhead.
From Implemetation Standpoint : We support HP solutions for the BI accelerator
Developed by SAP and Intel, the BI accelerator speeds up the performance of SAP NetWeaver BI installations and enables faster query response times.
HP has developed an out-of-the-box solution for the BI accelerator appliance. This solution is based on 64-bit Intel® Xeon™ Processors, low-cost HP ProLiant blade servers, and HP StorageWorks Enterprise Virtual Array (EVA) technology.
The HP solution for BI accelerator provides a number of benefits for our customers:
Our engineering team optimized the HP reference configuration for maximized and predictable storage and network I/O
Easy-to-order, fixed price installation service
On a global level, we offer factory pre-installation and pre-integration of all BI accelerator components before delivery.
Quick entry into operation
The box solution ensures an optimum fit into any existing SAP NetWeaver BI environment without complex and costly integration or migration efforts.
We have established standardized and globally consistent BI accelerator support processes for operational stability and availability.