Base architecture selection based on BI application

The amount of data used for SaaS applications is facing a TB-level growth rate, with different SaaS applications, and not exactly the same data structure, with text, graphics, and even small databases. With the distributed characteristics of the cloud service platform, the SaaS application data can be distributed on different servers, and how to data mining the heterogeneous data is a difficult problem faced by the enterprises in cloud era.

Challenges faced by enterprise data mining in cloud era

Mining efficiency: After entering the cloud computing era, BI's thinking has been transformed. Previously based on closed enterprise data mining, and faced with the introduction of Internet applications after a large number of heterogeneous data (it is expected that by 2020, the explosive growth of data will break through 35ZB (1ZB=10 billion TB)), the current parallel mining algorithm is very inefficient.

Multi-source data: After the introduction of cloud computing, the location of enterprise data may be in the provision of public cloud services platform, but also in the enterprise's own private cloud, how to face different data sources for mining is also a challenge.

Heterogeneous data: The most important feature of web data is semi-structured, such as documents, reports, Web pages, sounds, images, video, and so on, and cloud computing brings a large number of SaaS applications based on Internet models, and how to comb effective data is a challenge.

The data mining of SaaS application hopes to improve the quality of data mining by introducing fast parallel mining algorithm through massive data storage platform.

How to choose a reasonable infrastructure

For enterprises, how to integrate a variety of application data mining, to extract the appropriate use of business information is a major urgent needs of enterprises. The traditional bi mode is based on data warehouse and is the model of relational database. In the face of the rapid growth of heterogeneous data, the traditional data warehouse and the original parallel computing technology due to low mining efficiency, can not solve the massive data mining work, affecting the timely extraction of data.

Business intelligence systems have long been built on the traditional SMP architecture of minicomputer. With the increasing performance of X86 platform in recent years, increasing usability and rapid expansion, the X86 platform has begun to erode the share of minicomputer in more and more market areas, and business intelligence has become another battlefield for X86 architecture to launch an attack on RISC minicomputer. For example, Oracle's Exadata Database cloud server, based on the Intel Xeon Platform, delivers high OLAP performance (data warehousing applications) and OLTP performance on a X86 architecture, with unique SmartScan technology and a design that moves down the data processing process. In addition, IBM has launched a business intelligence solution based on the X86 platform, based on IBM's unique EX5 architecture server and XIV Grid storage System provides intelligent information processing capabilities that are not delivered to minicomputer.

Purchase points:

1, High availability: BI's infrastructure layer, the need to establish a data mining cloud service platform, and this platform is necessarily highly available.

In terms of high availability, you need to focus on three aspects: first, data protection, the need to use CRC, ECC and other hardware mechanisms to the transmission of data validation, error correction, if unable to correct, the damage to the data to be isolated to ensure that no more data to avoid the system restart and downtime.

Current Intel Xeon 7500 or E7 solutions have many advantages, such as low cost, high performance, high reliability (RAS), and scalability. On the scalable performance, the X86 platform extends horizontally outward, which consists of more than two machines forming a cluster. Meet the load requirements of most enterprise-critical application environments, including databases, business applications, and virtualization with higher memory and CPU requirements. In order to avoid the traditional UNIX dual-computer scheme "expensive, standby resources in peacetime seriously idle waste, host failover during the user Service forced to pause" and many other difficulties.

In addition, 75,001 of designs have minimized planned downtime, including system partition management technology, thermal addition of CPUs and memory, and thermal removal, minimizing system maintenance time.

2, Virtualization: Data Mining cloud services or to rely on virtualization technology, to calculate the allocation of resources and scheduling, that is, virtualization technology is the data Mining cloud service technology support.

Don't be fooled by the concept.

Large data has many different uses. Therefore, enterprises need to use different data mining platform according to their own business situation. For those customers who focus on application analysis and processing requirements, there are a number of specialized solutions, such as HP Vertica, plus a lot of high-performance NAS or target systems.

Similarly, for video, security surveillance, CCTV, simulation, bandwidth, or throughput, consider HP Ibrix, Dell Exanet, BlueArc, HDS, NETAPP, Data Direct nx, Oracle 7000, EMC Isilon and Vnx.

In general, users may face a whole host of market hype that will persuade you to move to more expensive systems. Maybe your current system is good enough--if it can be extended, what the vendor offers you won't necessarily run well in your current environment.

For users, you need to be wary of all sorts of hype about big data, and they might want to narrow your options. In addition to the opportunities that large data can bring, there are many different aspects to consider, such as its characteristics, applications, usages, and deployment scenarios.

(Responsible editor: The good of the Legacy)