Server knowledge details

Http://www.myrui.cn/article.asp? Id = 290

I. Main Performance Characteristics of servers

The English name of a server is "server". It refers to a special dedicated computer that provides various services for the client in the network environment. In the network, the server is responsible for data storage, forwarding, publishing, and other key tasks. It is an indispensable part of various Client/Server (C/S) mode networks. In fact, server hardware is not strictly required. Especially for medium and small enterprises, their servers may be a PC with good performance. The difference is that they have installed a dedicated server operating system, therefore, such a PC can assume the role of a server, commonly known as a PC server, to complete various required server tasks. Of course, because the performance gap between a PC and a dedicated server is far from that of a dedicated server, we can imagine that the servers played by the PC are in the network connection performance, or other aspects such as stability cannot undertake high-load tasks, and can only be applied to small networks with simple tasks. This article and subsequent articles do not introduce such PC servers, but specialized servers.

However, after all, the server is actually a computer, which is also developed by a PC. In the early days when the network was not very popular, there was no such name as a server. At that time, there were only large computers and micro computers in the computer field. However, with the development and popularization of networks, especially local area networks, computers at the intermediate level of "servers" began to be accepted by the industry and continued to develop with the popularization and development of networks. Even so, the servers are not exactly the same as the computers we normally see. Otherwise, few of the many server vendors in China can actually produce medium and high-end servers, even the world has only a few such strengths as IBM, HP, and sun. Dell can only produce some medium and low-end servers because of the special requirements of servers, this is the four main features of the server (usually called "four features "). Although servers are similar to PCs, such as mainboards, CPUs, memory, and hard disks, these hardware is not used by General PCs and are specially developed for server environments, although it looks similar. Because of this, the price of servers is usually very high. The mid-range servers are tens of thousands of yuan, and the high-end servers are dozens or millions. Of course, we have also seen many brand-name servers with a price of just a few thousand yuan, such as Dell and HP. However, these servers are all entry-level servers. In terms of performance, they are only equivalent to a high-performance PC, which can be called a "PC Server ", this was developed to meet the needs of some small enterprises for dedicated servers. For this reason, these servers only have a small number of server performances.

With the continuous development of PC computer technology, there are some abnormal phenomena between server and PC technology. Originally, PC technology has always been lagging behind. Many PC technologies are transplanted from servers, but now there are some changes. Many PC performance has been greatly improved, such as high CPU clock speed, MHz bus frequency, Sata serial disk interface, PCI-Express interface, and hyperthread technology, these new technologies have never been used for servers, and their corresponding performance is better than the original server performance. Therefore, these technologies will soon be widely used in the latest servers. Of course, the server still has many advanced special performances.

as a server, the first requirement is that it must be reliable, that is, "availability ". Because the server is facing the users of the entire network, rather than the users logged on from the local machine, the server cannot be disconnected as long as there are users in the network. In some special application fields, even if users do not use some servers, they have to work continuously because it must continuously provide connection services for users, whether at work or off work, this is also the root cause for the high stability required by servers, whether on work days, off holidays, or holidays. Generally, dedicated servers require uninterrupted work, especially some large network servers, such as servers used by large companies, website servers, and web servers that provide public servers. These servers may only work once, that is, once they have been purchased, fully installed and configured, and put into use until they are completely decommissioned. If the problem persists, can such a network continue to work normally for a long time? This is the most critical performance of the server, and it is also a prerequisite for assuming the role of the server, even a PC.

In addition, if the server is to provide services for so many users, there is no high connection and computing performance to bear. This means that the server is "available ". We usually feel slow when we use a single machine. If the server is the same as the PC we usually use, how can we get the computer's response and completion for so many requests? Therefore, the performance and speed of servers are quite different from that of normal PCs. In order to achieve high speed, servers are generally guaranteed by Using symmetric multi-processor installation and inserting a large amount of high-speed memory, in this way, the hardware configuration of the server is essentially different from that of a common computer. Its motherboard can be installed with several or even dozens or hundreds of servers (for example, sun's fire 15 K can support up to 106 CPUs. These CPUs are exactly the same as those in general PCs. We know that the most important parameter for a common CPU is the clock speed. The higher the clock speed, the faster the calculation speed, but it is far from the same in the server CPU. Generally, the server CPU frequency is relatively low, for example, Intel's server CPU clock speed is usually around P4 2.0ghz, which is far lower than the PC's CPU speed of GHz, while other brands have lower CPU clock speed, however, these server CPUs have excellent computing performance. The higher the CPU clock speed, the higher the heat emitted during operation, which brings the biggest unstable factor to the server. On the other hand, the higher the server computing performance, not only through the increase of the clock speed, it is generally enhanced by other parameters. In addition, most medium and high-end servers can greatly improve the overall computing performance of servers through symmetric multi-processor systems, there is no need to improve the computing performance by clock speed in a single CPU. Note that the number of CPUs on the server must be double, that is, the so-called "symmetric multi-processor system ". The same is true for memory configurations, which are fundamentally different from the memory used by General PCs in terms of memory capacity, performance, and technology.

In addition, the server must have certain "scalability" because the network cannot remain unchanged for a long time. If there is no scalability, when the number of users increases, the number of servers is tens of thousands, even if hundreds of thousands of servers are to be eliminated in a short period of time, many enterprises cannot. To maintain high scalability, you usually need to have a certain number of scalable space and redundant parts on the server (such as disk matrix bit, PCI and memory slot ). Of course, the hardware configuration is far more than that. We will introduce it in the following sections.

In terms of the main features of the server, there is also an important aspect, that is, the server must have a certain degree of automatic alarms, and have corresponding redundancy, backup, Online Diagnosis and recovery systems, in case of a fault, the server can be recovered in time, which is "manageability ". Although we say that the server needs to work continuously, even a good product may fail for a day, people often say that it is not unknown that it may be bad, but does not know when it will be bad. Although the server has sufficient stability assurance, what should we do if a fault occurs? If we stop the service like a computer we usually use, it is impossible for a large server, in this way, the entire network may be paralyzed and the loss cannot be measured by money. Server manufacturers have put forward many new technologies to solve this problem, such as redundancy technology, system backup, Online Diagnosis Technology, Fault Pre-alarm technology, memory check and Error Correction Technology, Hot Plug-in technology and remote diagnosis technology, this allows the vast majority of faults to be repaired in a timely manner without downtime.

The server mentioned above is "four features", "scalability, availability, manageability and availability", that is, the server "Suma" we often see ". The "four features" are very complicated. We will introduce them in the next article.

Ii. Main server exterior features

The above describes the main features of the server in terms of performance. These performances can only be swallowed up by a reader who has just been in contact with the server. This is not surprising, because the server, the most technically complex network product in the computer network, cannot be clearly explained through the above simple text. What's more, the main features of the servers mentioned above are not easy for many readers to access. Are there some external performances that are easy to understand on the servers? Yes, and it is very obvious. The following describes some of the main features of the server in terms of appearance, so that you can have a clearer understanding of the server, which is very easy to understand. The following are the Appearance Features of the server:

1. Large Chassis

In terms of appearance structure, the server chassis is generally relatively large. Some servers, although seemingly similar to ordinary PCs, are larger, even for entry-level PC servers, 1 The figure on the left shows the Dell poweredge 400sc entry-level server (currently only $5808 ). For some medium-and high-grade professional servers, the difference between the chassis is even greater, and some are a high cabinet. 1 The picture on the right shows the S80 high-end server of the ibm rs/6000 series.

Of course, the large server chassis is not intended to mislead consumers, but it has a big reason. On the one hand, because the server needs to be installed and connected to many devices, need to occupy a large space, but also prepare some backup equipment placement, such as tape drive, disk array, multi PCI or PCI-X plug-in card, etc, it also takes up space. Servers usually need to be connected to many devices. For example, many adapter cards need to be inserted into the server motherboard or dedicated board, some also need to install several hot plug-in hard disks and power supplies (commonly known as "redundant power supply"), and generally there is a tape drive for system backup and recovery. In addition, due to the installation and connection of many devices, the heat is also very large during work, so there must be enough space for heat dissipation to ensure that the server can work stably for a long time. All of the above determine that the server's chassis architecture must be larger than ordinary PCs. Even if some media say diy pc servers, in order to ensure long-term stable operation of servers, we must increase the chassis, be sure to pay attention when purchasing or configuring PC servers.

Of course, the above mentioned large chassis is only for integrated tower servers. With the refinement of server applications in recent years, major server vendors have launched dedicated functional servers for a variety of specific applications, such as server products launched for space-Intensive Environment applications to rack servers, blade servers are specially optimized for space occupation, greatly reducing the space occupied by the chassis. 2 shows a blade rack functional server.

2. Large disk and memory capacity

This is also easy to understand. servers must meet the needs of many users and accept requests from all users. Moreover, they must install and save many large-capacity server-dedicated systems and software, as well as some other database files, this requires the server's hard disk capacity to be large enough. In the past, because the hard disk capacity was relatively small (only a few hundred megabytes in the past), the disk matrix is usually used, and many disks are installed in parallel on the disk shelf of the server, although this is not only used to increase the disk capacity of the entire server, it was also one of the main purposes of increasing the disk capacity at that time. At present, the hard disk capacity has been greatly improved, and the maximum size is more than GB. Therefore, the average small and medium enterprise network server requires only one hard disk, the main purpose of using the disk matrix is to improve disk access performance and secure recovery. Of course, for large network servers, such as some portal servers, the disk capacity cannot be met by a hard disk at present, because the disk capacity required for such servers is usually Tb (1 TB = 1000 GB), multiple disks or disk matrices may be used. In addition, in order to improve the disk access speed, the server hard disk usually uses the SCSI interface, and the speed of the hard disk is more than 10000ppm.

In terms of memory capacity, we mainly take into account the user access speed requirements of the server. We know that the internal existence largely determines the system's operation speed. The larger the server network, the more complicated the data traffic, more memory is required. Currently, most servers for small and medium-sized enterprises are larger than 1 GB, and some high-end servers can support up to TB of memory capacity. Of course, the server's memory requirements are far more than capacity requirements. There are special requirements for memory access speed and Error Correction performance, which will be described later.

3. Large Motherboard

Generally, the server motherboard is much larger than the PC motherboard, mainly because it requires more components than the PC, such as more PCI (more than 5), PCI-X, memory slots (more than four), there may be multiple CPU sockets. 3 shows the dual-CPU socket server motherboard of an i875p chipset of Asus. If it is a motherboard that supports 4 or more CPUs, It is larger. Some of the more concentrated functional parts are provided with a separate board to save space on the motherboard, if some servers PLACE all PCI or PCI-X slots in one board and rotate them in a separate box outside the chassis, they connect to the motherboard through a data cable.

Of course, this is also for common integrated servers. For blade-type functional servers, the main board may not be large. Note that it is not absolute.

4. Are there Useless components?

In some of the more upscale servers, we can see some strange phenomena, that is, how can there be two power supplies and two fans in one machine, there are also some NICs that are not actually connected, and of course there are hard disks that seem useless. In fact, none of these are useless, but they are currently not used. They are used to take over the work when the same part is working fails, commonly known as "redundant parts ". With these redundant parts, even if these parts are faulty (some even replace them when the system finds some abnormal parts before the formal failure occurs ), this does not cause the server to stop working and the entire network to continue normal operation. This is very important to ensure the high stability of the server. Power supply, fan, and nic redundancy may be easy to understand. Just replace them directly. For Hard Disk redundancy, hard disks used for redundancy also need to be regularly backed up on working hard disks, only in this way can the redundant hard disk be replaced immediately after the latest network system configuration to provide services.

5. Support hot dial

Although hot plug-in is a technology, it can still recognize and recognize this technology through our usual operations, because the devices using this technology are usually expensive, so of course this feature is not available in PC. Hot swapping is mainly used to facilitate the maintenance of servers (which is also necessary for servers). If you find that the hard disk is insufficient or a hard disk is damaged, if the hard disk supports hot dial, we only need to insert the new hard disk into the reserved location of the server, or directly dial the bad hard disk from the server for repair, all these operations do not need to shut down the server system, which ensures uninterrupted operation of the server. Currently, Hot Plug-in technologies are supported, including hard disks, power supplies, fans, and PCI adapter cards (mainly network cards. At present, some high-end servers support hot plug-ins for memory and CPU. All servers that support Hot Plug-in technology can be online (do not turn off the server power) in this case, you can directly Insert new parts or dial the old parts from the server, which greatly facilitates the maintenance of the server and ensures that the server remains running for a long time.

Iii. Classification Based on Different Application Levels

By application hierarchy is also known as "by server level" or "by network scale". It is the most common way to divide servers, it is mainly divided based on the application level of the server in the network (or the server level. It should be noted that the server grades mentioned here are not divided by the CPU clock speed of the server, but measured based on the overall performance of the entire server, especially by some server-specific technologies. Servers can be divided into entry-level servers, working-level servers, department-level servers, and enterprise-level servers.

1. Entry-level server

This type of server is the most basic type of server and the lowest type of server. With the increasing PC technology, many entry-level servers have similar configurations with PCs. Therefore, some people think that entry-level servers are equivalent to PC servers.

This type of server does not have many server features, and generally only has the following features:

· Some basic hardware redundancy, such as hard disks, power supplies, and fans, is not necessary.
· Hard disks with SCSI interfaces are usually used, and SATA serial interfaces are also used.
· Some components support hot plug-ins, such as hard disks and memory, which are not required.
· There is usually only one CPU, but it is not absolute. For example, sun's entry-level service can support up to two processors.
· The memory capacity is not very large, generally within 1 GB, but usually the server dedicated memory with ECC error correction technology is used.

This type of server mainly uses Windows or Netware network operating systems, which can fully meet the needs of small and medium-sized office network users for file sharing, data processing, Internet access, and simple database applications. This type of server is very similar to a general PC. Many small companies simply use a high-performance branded PC as the server, the price is almost the same as that of a high-performance PC brand. For example, the price of Dell's latest poweredge4000 SC is only 5808 yuan, and HP also has entry-level servers with similar configurations and prices.

The number of terminals connected to entry-level servers is limited (usually about 20). Besides, the stability, scalability, and fault tolerance redundancy performance are poor, it is only applicable to small enterprises that do not have large database data exchange and daily work network traffic and do not need to start up continuously for a long time. However, it should be noted that there are currently some large server development and production vendors which will be divided into several levels in the enterprise-level servers we will talk about later, the lowest-level enterprise-level server is called an "entry-level enterprise-level server". The entry-level server mentioned here does not have the same meaning as the "entry-level server" mentioned above, however, this division is still relatively small. Another point is that such servers generally use intel's dedicated server CPU chip, which is based on the intel architecture (commonly known as the "IA structure"). Of course, this is not a hard standard, this is due to the application level requirements and price limits of the server. Figure 1 shows an old-fashioned Sun Enterprise 2 entry-level server. In terms of appearance, it is no different from a common PC. in figure 1, the HP Server tc2110 is shown in the right figure, and its chassis is a general vertical chassis.

2. Working Group Server

The workgroup Server is a level higher than the entry-level server, but still belongs to low-end servers. It can also be seen from the name that it can only connect to one working group (about 50) so many users, the network size is small, the stability of servers is not as high as the application environment of enterprise-level servers as we will talk about below. Of course, the requirements for other performance are also lower. Working group servers have the following main features:

· Generally, only single-or dual-CPU application servers are supported (but not absolute, especially Sun's workgroup servers, which can support up to four processors, of course, the price of this type of server is somewhat different ).
· Supports large-capacity ECC memory and an SM bus that enhances server management.
· Comprehensive Functions, strong manageability, and easy to maintain.
· Use Intel Server CPU and Windows/Netware network operating systems, but some of them adopt UNIX operating systems.
· It can meet the needs of small and medium-sized network users for data processing, file sharing, Internet access, and simple database applications.

Compared with entry-level servers, Working Group servers have improved performance, enhanced functions, and certain scalability. However, the fault tolerance and redundancy performance is still incomplete and cannot meet the needs of large-scale database systems, however, the price is much higher than the former, which is generally equivalent to the total price of 2-3 high-performance PC brand machines. 2. The HP lc2000 workgroup server is shown in the figure on the left. This series of servers are designed for the computing needs and budgets of small businesses. The performance and scalability enable these servers to meet the needs of applications, such as files and printing, e-mails, order processing and e-trade are extended. Figure 2 the figure on the left shows Lenovo's newest 10-Gigabit T200 2100 workgroup server. It uses a dual Intel Xeon 533mhz processor and the latest SATA technology in a PC; the right figure shows HP's ProLiant ml350g3 workgroup server, which can use 1-2 intel GHZ-kb Xeon server dedicated CPUs and support hyper-Threading Technology.

3. Department-level servers

This type of server is a mid-range server. Generally, it supports a dual-CPU or higher symmetric processor structure and has complete hardware configurations, such as disk arrays and storage bays. A department-level server is characterized by a large number of Monitoring and Management circuits integrated in addition to all servers of the Working group server, providing comprehensive server management capabilities, it can monitor status parameters such as temperature, voltage, fan, and chassis, and use standard server management software to keep managers informed of the server's working conditions in a timely manner. At the same time, most department-level servers have excellent system scalability, which can meet the needs of users to promptly upgrade the system online when their business volume increases rapidly, fully protecting users' investment. It is a necessary link for the scattered grass-roots data collection units in the enterprise network to maintain smooth connectivity with the highest-level data centers. It is generally the first choice for medium-sized enterprises and can also be used in finance, post and telecommunications industries.

Department-level servers generally use the CPU chips developed by IBM, sun, and HP. These chips are generally in a hierarchical structure and generally use UNIX operating systems, linux is also widely used in department-level servers. In the past, only IBM, HP, sun, and Compaq (now integrated into HP) vendors were able to produce department-level servers. However, with the development of other server vendors, there are more vendors that can develop and produce department-level servers than before. Several companies in China also have this strength, such as Lenovo, sugon, and inspur. Of course, because there is no industry standard to define what type of server configuration can be regarded as a department-level server, many companies that are not strong also claim that they own department-level servers, however, the product configuration is basically no different from that of the entry-level server. You should pay attention to it.

Department-level servers can be connected to around 100 computer users, suitable for small and medium-sized enterprise networks with high processing speed and system reliability. Their hardware configurations are relatively high, its reliability is higher than that of working group-level servers. Of course, its price is also higher (usually the total price of about 5 high-performance PCs ). Because many components need to be installed on these servers, the chassis is usually large and rack-based. Figure 3 shows IBM @ server xseries 230 department-level server products. The appearance of this server is quite different from that of an ordinary PC. Figure 3 shows the Dell poweredge 4600 department-level server, which can be installed with a maximum of two Intel Xeon processors ranging from 1.8 GHz to 3.0 GHz. Its chassis is not cabinet-type, but it is also larger than the general entry or working group server chassis.

4. enterprise-level servers

Enterprise-level servers belong to the ranks of high-end servers. As a result, there are not many enterprises that can produce such servers, but also because there is no industry standard hardware that specifies the level of enterprise-level servers to be achieved, so now we can see that many enterprises that do not have the development or production level of enterprise-level servers claim that they have enterprise-level servers. Enterprise-level servers use at least four or more CPU symmetric processor structures, and some have up to dozens. In addition, it generally has an independent dual-PCI channel and memory expansion board design, with high memory bandwidth, large capacity hot swapping hard disk and hot swapping power supply, strong data processing capabilities and cluster performance. This type of enterprise-level server has a larger chassis, which is generally cabinet-type, and some are composed of several cabinets, just like a mainframe.

In addition to all server features of department-level servers, enterprise-level server products, it also has a high degree of fault tolerance, excellent scalability, Fault Pre-alarm function, Online Diagnosis and Ram, PCI, CPU, and other hot plug-in performance. Some enterprise-level servers also introduce many excellent features of large computers, such as the enterprise-level servers of IBM and sun. The chips used by these servers are also dedicated CPU chips developed by several major servers and developed by the manufacturer. The operating systems used are generally Unix (Solaris) or Linux. At present, there are only a few manufacturers that can produce high-end enterprise-level servers worldwide, such as IBM, HP, and sun. The enterprise-level servers of most domestic and foreign manufacturers can only be regarded as medium and low-end enterprise-level servers. Enterprise-level servers are suitable for financial, securities, transportation, post and telecommunications, communications, or large enterprises that require massive data processing, high processing speed, and extremely high reliability.

Enterprise-level servers are used in large networks with hundreds of computers connected to the Internet and high processing speed and data security requirements. Enterprise-level servers have the highest hardware configuration and the highest system reliability. 4. the ibm rs/6000 S80 enterprise-level server is shown in the figure on the left. It is the first Rs 6000 platform model with rs64 ⅲ microprocessor, its multi-processor system can support up to 24 symmetric processors, and the chip is based on IBM's excellent copper technology, making the processor faster and more reliable. As shown in figure 4 on the right, sun's fire TM 15 K high-end enterprise-level server product supports up to 106 ultrasparc III Cu 1/2-MHz symmetric processors with a memory capacity of TB.

we have introduced the general classification of servers. Note that the boundaries between these four types of servers are not absolute, in addition, with the development of server technology, server technologies at various levels are constantly changing and developing. Maybe only a department-level technology will also be available on entry-level servers one day in the future. In addition, these types of servers do not have a hard standard in the industry to strictly classify them. Most of them are for the entire server product line of different manufacturers. Because there are many server models and hardware configurations are quite different, users do not have to stick to a certain-level server, but should follow their network scale and service needs, it is important to consider the relative redundancy and system scalability. As an enterprise expands, its requirements for servers will also grow. If the server has a strong scalability, then there will be no re-purchasing servers in a short period of time. We only need to buy some extended parts to complete the server performance upgrade. For example, the enterprise-level X-architecture server of IBM has a concept of "On-Demand scaling" or "Pay as you grow (Pay as you grow, to fulfill your wish. In fact, there are similar ideas and HP's "pay-as-you-go" philosophy. Of course there are still large differences between these two concepts, IBM's "pay-as-you-go" means that you can choose not to purchase server components that will be available in the future. In the future, you need to purchase additional services, this reduces the investment risk of enterprises for devices. While HP's "pay-as-you-go" means that HP sells you the entire set of server devices first. If you think that some components are currently not used, and may be used by your organization in the future, at present, your organization only needs to pay part of the cost you currently use. If you do not need to use it for the time being, you can not pay for it first. When you think you can use it, you can install it for you and pay for it again, in fact, the same purpose as IBM's "pay-as-you-go" above is to reduce the customer's device investment risk.