On the application of dual-core server technology: Technical Articles

What is dual-core technology, I believe most people can understand, but why use dual core? What benefits can it bring to us? How to use to play the true power of dual-core? These questions still lack a more definitive answer. In fact, several professional media in China have also published some technical articles to discuss this topic, but a little DIY to analyze the taste of dual-core. We feel that the dual-core processor on the personal computer application value and popularization speed should be far behind this technology for the server domain, so today I am completely in the position of server application, combined with some of their views, to talk about dual-core technology:

What is dual-core technology?

&http://www.aliyun.com/zixun/aggregation/37954.html ">nbsp; Using a more acceptable version, a dual-core processor is a processor core that has two functions on one processor based on a single semiconductor, but it's hard to explain essentially why a dual-core is born, and if you want to pursue the origins of the two cores, you have to start with some basic knowledge of physics and electronics:

Unable to overcome power and calorific value

Now the chips are made on a single crystal wafer, large area of the wafer itself will inevitably have some defects, the process of production sometimes will also bring defects in silicon crystals. In addition, although the factory cleanliness is very high, but in the production process or unavoidable to bring some impurities, interconnect layer in the manufacturing will also introduce their own defect form. Such a large scale chip is hard to guarantee that there is no defect at all. Now that the semiconductor is in the process of 90 nm, the effect of defects and impurities is more common, the size of the transistor has been and some defects or impurities in the size of a magnitude, even smaller transistor size. In general, non-fatal defects can lead to reduced circuit performance, in order to compensate for the loss caused by the defects, it is necessary to give the circuit higher voltage and greater current, many defects in reducing the transistor performance while also increasing the power of the transistor, the previous CMOS digital circuit is the main power consumption is the dynamic power consumption, But after the 90nm process, the situation has changed a lot, many of the original basic negligible power consumption factors now occupy a larger proportion, such as circuit interconnection loss, leakage loss. The impact of leakage loss has increased dramatically. That's why, after the 0.09 micron process, the heat and power consumption of the various processors is the cause of the upward trend, high-frequency Prescott processor power consumption will be nearly 100W, you know, a solder can quickly melt the soldering iron is only about 50W, therefore, high power consumption and the frightening calorific value of the CPU to continue to advance to a major obstacle, at that time there is a very Many people predict that after the birth of the 4GHz processor, you will be able to boil the eggs through the CPU heatsink.

Solutions beyond HF

Since it is not possible to achieve higher frequencies by continuing to increase the frequency, how should the user's need for product performance improvement be met?

The first approach, of course, is to use 64-bit technology, why the 64-bit operation is more powerful than the 32-bit performance, this is a good understanding, for example, you use 32-bit CPU, to export a 64-bit data, you need to consume two time pulse, but if you use a 64-bit CPU, as long as one time pulse is done.

Another approach is to use the multipath symmetry technique, the basic theory of course is to use two processors to do a processor, like disk array technology, RAID 0 is the use of two hard drives to access a file at the same time, each access half, so spend half the time, if the use of 4 hard disk, That is, each block access 1/4, the time required is naturally shorter; the CPU's multipath symmetry is also the case, therefore, the high-end server many use 4-way, 8-way equipped, is to improve product performance.

The difficulty of multi-path symmetry

It is not difficult to achieve high-performance array storage with 2 or even 4 hard drives, because the RAID controller automatically controls how the hard drives work, and the programmer or CPU simply tells the RAID controller what data to read or write, and how to split the data. Several hard drives how to allocate read and write problems do not need computer users, programmers to consider, so the original operating system and software do not need to make changes, but the CPU's multipath symmetry work is different, because all the computer instructions are CPU to execute, If the programmer only orders the CPU to execute what instructions and not tell the computer which CPU to execute the instruction, then there will be chaos, and if it is two CPUs, the command scheduling allocation is also better design, if it is 4-8, then the situation is very complex, like two elevator joint control program easy to write, But the joint control procedures for 4 elevators are very difficult to write, because they are 2 of 2 times and 4 of 2 times the relationship.

In addition, because multi-core chips integrate two or more independent CPU cores into a single wafer, some people suspect that their fatal weakness is data movement between these CPU cores, and there are some performance bottlenecks when multi-core systems run different applications competing for the same I/O path. But so far, the impact of these bottlenecks has been small.

How to use a lot of road symmetry

The multiple symmetric mode of work requires programmers to design the software itself can support a number of CPU command scheduling, so that all the CPU is the most reasonable and most efficient scheduling to work together. Because most of the application software is the use of a large number of API functions developed, so as a command-machine interaction role of the operating system is the fundamental and key to the development of multiple symmetric, a set of operating systems can be a good coordination between the working relationship between multiple CPUs, is a multi-channel model can successfully play its performance key

Microsoft's server-level operating system supports multipath mode, and the latest 2003 server also improves the multipath-symmetric command-scheduling design, enabling the server to perform better in multipath mode. In addition, the design of the application itself is also a determining factor in the effectiveness of the multiple-symmetric work, for example, the 3D games are now based on single-threaded development, which means that the game does not invoke the operating system's multithreaded functions, so that even with a multiple-symmetric operating system, the application can not be improved , therefore, you can note that so far the multi-channel design is only in the server, workstations and other high-end areas, because from the cost and effectiveness of the use of personal consumption in the area of multiple design is not wise.