Example Analysis Teach you how to choose the database server

The database server is the core of the business system, with large business volume and large amount of stored data. It is responsible for the storage and processing of business data, so the choice of critical database servers is particularly important. Server reliability and availability are primary requirements, followed by data processing capabilities and security, followed by scalability and manageability.

Depending on the type and size of the application, the performance requirements of the database for the server are different. For large databases (ERP, OLTP, data mart), servers are often used only to run databases, or to run only a single application. The capacity of the database is above 1TB, which requires high CPU processing capacity. The large-capacity memory is a data cache service and requires good IO performance. When using such applications, it is usually necessary to have a higher CPU clock speed. So, how to choose a database server for a certain industry or even a certain project?

Five principles for database server selection

First, database server selection should follow the following principles

1) High-performance principle

Ensure that the server you purchase can not only meet the needs of the operation and business processing of the operating system, but also meet the needs of business growth in a certain period of time. Generally, the required server TpmC value can be calculated according to the empirical formula, and then the TpmC values ​​published by each server manufacturer and the TPC organization are compared, and the corresponding models are selected. At the same time, the market price/quote of the server is used to remove the calculated TpmC value of the unit TpmC value, and then select the server with high performance price ratio.

2) Reliability Principles

The reliability principle is the primary consideration in all selected equipment and systems, especially in large, long-running systems with large processing requirements. Considering the reliability of the server system, not only the reliability or stability of the single node of the server, but also the overall reliability of the connection between the server and the related auxiliary system, such as network system, security system, remote printing system, etc., should be considered. When necessary, you should also consider clustering technologies for critical servers, such as dual-system hot backup or cluster parallel access technology, or even possible full fault-tolerant machines.

For example, to ensure that the system (hardware and operating system) can operate normally (including repair time) in 99.98% of the time, the downtime should not exceed 0.5 hours in six months. The server needs to run continuously for 7×24 hours, which requires high security and reliability. The system has a mean time between failure (MTBF) of no less than 80,000 hours. If there is CPU damage or other mechanical failure, the server can automatically replace the work by the standby CPU and machine within 20 minutes, without human operation, to ensure data integrity.

3) Extensibility Principles

Ensure that the server you choose has excellent scalability. Because the server is the core of all system processing, it requires large data throughput rate, including: I /O rate and network communication rate, and the server needs to be able to handle the amount of data brought by a certain period of business development, the server needs to be able to respond at the corresponding time Upgrade the user's own needs according to the needs of business development, such as CPU model upgrade, memory expansion, hard disk expansion, network card replacement, number of terminals, mounted disk arrays, or cluster systems that form concurrent access to centralized data with other servers. Wait. This requires the server to be purchased to have a good room for expansion as a whole. General database and billing application servers use clustering to increase reliability in the design of large billing systems. The attached disk storage system can be implemented by DAS, NAS or SAN according to data volume and investment considerations. .

4) Security Principles

The server processes most of the core data of the relevant system, where it stores and runs key transactions and important data. These transactions and data are an important asset to the owner and their security is very sensitive. The security of the server is inseparable from the overall security of the system, such as security of the network system, data encryption, and password system. The server needs to be designed in its own right, including hardware and software. From the perspective of security, it is necessary to ensure its own high security.

5) Manageability Principles

The server is both a core and a node part of the system as a whole. Just as the network system needs to be managed and maintained, it also needs to effectively manage the server. This requires the server's hardware and software to support the standard management system, especially the operating system on it, as well as some important system components.
Example explanation database server selection

In order to let everyone have a clearer understanding of the above principles, let us elaborate on the database in the data security server of a provincial data center exchange area of ​​Jinbao Engineering. The method of server selection.

The database server in the data layer server of the provincial data center exchange area as a social insurance relationship, the database server for remote data exchange and ex-gratia medical data exchange, and the exchange of information on the social insurance relationship of the incumbent And exchange of pension-related information (such as basic conditions of personnel, payment standards, living conditions, etc.), while preserving death information and public service information, temporarily buffering macro-decision reporting data and fund supervision information. Considering its role as the hub for data exchange between the central, provincial and municipal data centers, the key to the supported applications should be high-end server systems. The specific configuration requirements are as follows:

1) Server processing capacity

In order to support the province's off-site transfer, medical treatment in different places and receiving pensions from different places, it requires high transaction data processing capabilities. The TPC is calculated as follows:

Assume that the total number of insured persons in the province is C=9.8 million, the average number of traders on the trading day is a1=1‰, and the number of database transactions per transaction is a2=5, then: daily actual transactions The quantity M= C×a1×a2; the concentrated trading time T=120 minutes on the trading day; the trading volume ratio Ct=80% in the trading day concentration period; the ratio of the benchmark TPC index value to the actual transaction value M0=6:1; CPU processing The capacity margin is M1=30%-45%, which is 35%; the annual processing capacity growth rate is P=30% within 3 years.

According to the empirical formula, TPC=(M×M0×Ct/(T×(1-M1)) ×(1+30%)3=89,435 is calculated. That is to say, server selection should be considered Use a high-end server system configuration with a TPC value of at least 100,000.

2) Memory capacity

Based on experience and similar traffic and environment, the memory capacity should be 1G/CPU×CPU, from At present, the indicators of mainstream hardware manufacturers, the TPC value should reach 100,000, generally need to configure 8 CPUs, so the memory is recommended to configure 8GB.

3) Bus I/O bandwidth

In the configuration of high CPU and large memory, the host system bus bandwidth and I/O bus bandwidth must be high. Otherwise, the system Performance will create a bottleneck.

4) Storage capacity

The average data volume of the exchange area is 164.8GB, and the peak data volume is 164.8GB×1.5, considering 0.2 times the database index and system footprint; after RAID protection 60 % storage utilization; future data growth, need to provide 30% of data expansion capacity and other factors, the total storage capacity is: 164.8 × 1.5 × 1.2 /60% /70% = 706GB, using Fibre Channel array in SAN as data storage .

5) Reliability, scalability, etc.

As a production database server that supports off-site business, it is a real-time service. The server system requires high reliability and reliability. Sex must be above 99.99%, providing 7×24 availability throughout the year, configured as a two-machine cluster. The system is designed with multi-component redundant structure, high-speed error checking and error correction memory, and monitoring and diagnostic functions.

Therefore, for the server selection, firstly, the system needs analysis of the business type and business complexity of the business system, and then the data capacity and data processing strength according to the requirements. Estimate and take into account the reliability, scalability, security, manageability and other aspects of the server to complete the final product selection.

New thinking: RISC or Nehalem-EX?

Due to the database server, especially the core database of key business areas such as financial telecommunications, the reliability and performance requirements of the server are very high. Database systems that have made such applications in the past often rely on RSIC minicomputers. Although the X86 server has a good price/performance ratio, it accounts for 95% of the total market, but there are still some core areas that have not been overcome.

However, in 2010, with the release of Intel's next-generation Nehalem-EX Xeon 7500 series server platform, this pattern is about to be broken. The Xeon 7500 has 8 cores, each core supports dual threads, and each processor can support up to 16 threads for parallel processing. Each processor has up to 4 QPI high-speed interconnects, which can be easily extended to 8-way systems. Third-party node controllers can be extended to more systems; each processor can share up to 24MB of L3 cache. With scalable memory buffering and scalable memory interconnect technology, each processor can support 16 memory slots, four channels. The server has a maximum memory capacity of 512GB DDR3. Moreover, Nehalem-EX introduces 22 RAS features and implements the MCA recovery function available on the IA64 for the first time on the Xeon platform, providing greater reliability. Whether in terms of performance, scalability and reliability, it has approached RISC and even surpassed certain indicators.

Therefore, with IBM, HP, Dell, Inspur and other vendors launching four-way and eight-way servers based on the Xeon 7500, core key users will have many choices. Below, we will also launch a series of articles to help interpret the application performance, product recommendations, solution cases, etc. of the Xeon 7500 in the database field.