Source:
http://proxy4.smth.org/pc/pccon.php?id=784&nid=101952&s=all
Five software ideas will be too fast
Calculation of Drapper Awards
■ Our reporter fat sea
-------------------------------------------------- ------------------------------
Editor's note: On November 9th, the National Natural Science Foundation of China and Microsoft Asia Institute jointly held in the Symposium held in Tsinghua University, the seminar invited Professor Yang Zhenning, the first Chinese Nobel Prize winner , Draper Prize, CHUCK THACKER, Dean and Chief Scientist of the Microsoft Asian Institute, Senior Vice President of Microsoft, Vice President of Microsoft, Ling Dali, Vice President of Microsoft Research, etc. Master's participation. The speech of the seminar is exciting, due to the limited layers, the reporter only reports on the most exciting speeches, this is -
Moore's law will be terminated by the "S Curve"
Draper Prize is known as the Nobel Prize in the Electronic Engineering. The winner Chuck Thacker explained in detail his unique view of Moore's law in the topic of "21st Century Computer".
He first said in the speech that many people have misunderstandings to Moore's law, often simply contributes that the calculation speed is doubled while the calculation is increased. In fact, Moore's law is very accurate, it is accurately predicted: every 18 months, computer storage capacity and transmission speed will increase by 1.58 times.
At the same time, Chuck Thacker also believes that Moore's law is not a strict sense of mathematics or physics law, but is just a discovery.
Since the discovery of Moore Law in 1967, computer storage capacity and transmission speed have been well verified that they have been raised to the initial 1 million times. However, the challenges faced by Moore Law began to appear. Whether the Moore's law continued to continue to persist, it has always been the biggest problem of the debate of the science and technology world.
Chuck Thacker believes that Moore's Law will be terminated by the "S curve". He believes that almost all natural development processes are in line with the universal "S curve", that is, all even is a long development process, will eventually slowly tend to stop, stopping in the top of "S". Moore's law is only a phased law, but the "S curve" is controlled.
From the physical level, Moore's law will also encounter "ultimate challenges": the atom is too large, the speed is too slow. You can't etch the circuit narrower than the atomic diameter of the base material, and it is impossible to transcend the transmission speed (i.e., electromagnetic field propagation speed). Moreover, higher speed means greater energy consumption, greater energy consumption means more arduous heat dissipation and cooling work, which will be solved in smaller spaces, will also cause Moore's steps to slow down. .
In a specific design, more and more transistors are increased in increasing spaces, and their increased workload and communication cost will increase geometric multiple. Design a large CPU usually takes 150 to 200 people's work team, working hours for about a year and a half to two years. When the number of transistors of the CPU chip increases, the number may continue to increase. To ensure that so many people work together, and there will be contradictions as little as possible, and the management cost will be very large.
At the same time, manufacturing costs will also increase, so that economics may not exist. It takes 50 million US dollars in the 1980s, and now the cost is up to 1.5 billion US dollars. Almost very subtle changes, mean huge additional investments, of course, also means that it is impossible to happen. Moore's law is not a four-sea IT Bible, some IT products do not follow Moore's law, like hard drives, displays, batteries, etc., they follow other laws. For example, the development of battery technology has enabled us to improve in energy density and cost, and the cost is basically decreased at a rate of about 10% per year. Of course, these contemptors are increasingly becoming the main cost of the entire computer.
Chuck Thacker finally believed that Moore's law continued for more than ten years, will then be terminated. Professor Yang Zhenning is different from him, Yang believes that Moore's law will still work for a long time, this time may be a three or fifty years.
Five software ideas will be too fast
Before the 21st century, computer software is largely to overcome hardware defects. In the 21st century, these hardware restrictions will not be present. Chuck Thacker said that in terms of capacity and speed, we all have powerful computers. We will start thinking about how to write software; at the same time, some of our software thoughts will also be outdated.
Chuck Thacker believes that the first outdated software thought is virtual memory (Virtual Memory). The original reason for proposing virtual memory is that real memory prices are relatively high, and there is only a piece of space as a virtual memory on the hard disk. But because the hard disk speed is limited, the speed of virtual memory is far less than true memory. Later, outside of this demand, virtual memory also solves the problem of insufficient address space and additional protection. But as the price is greatly reduced, the virtual memory lost the main role. At the same time, because the hard disk speed becomes the bottleneck of the entire system, it also leads to the "chicken rib" in the virtual. However, the main operating system still retains the outdated design of virtual memory, although users can close it. However, in modern programming languages, the use of virtual memory has become less and less.
Chuck Thacker believes that the second outdated software idea is a directory file management system (Hierarchical File Systems). The original purpose of this design is to make users improve the management efficiency of documents or information, so that users have a way to meet routine. However, it has now caused serious problems. For example, only 45,000 files are required to install Windows and Office, which are distributed in a complex directory, and users cannot quickly find some files they need. Now, a person uses this structure to manage documents and information. After 6 months, he will be overwhelmed in the rapidly increasing documentation and information ocean. Chuck Thacker believes that it is best to use a full-text search management method. That is to say, introduce search techniques on the Internet to personal computers. At present, the desktop search for Google and other search engine providers are a good example.
Chuck Thacker believes that the third outdated software think is to verify the software quality (Verifying Software Quality By Testing). The parameters in modern software are increasing, and some software is more than 10,000, the software source program code exceeds 100,000 lines. It is more and more difficult to find all the bugs through the test. At the same time, because the software is huge, the problem is often concurrency, and each bug may have a chance to produce a fatal disaster. Chuck Thacker said that this requires more standardized methods, such as TLA launched by Silicon Valley Researchers, which combines a specification language with model inspection, which is more effective. Intel often uses this method when designing new protocols, which saves them a lot of money. Chuck Thacker believes that the fourth outdated software think is a low-level programming language (low-level programing language). Low-level programming languages (such as machine language and assembly language) can freely call memory resources, which leads to many system unstable problems. If you use advanced programming languages such as Java or C language, you can automatically manage memory resources, making many frequent errors that have previously occurred. At the same time, programmers will not encounter problems like hardware details when using low-level programming languages, making the BUG rate of the software reduced.
Chuck Thacker believes that the fifth outdated software idea is a single-threaded program. In a single processor system, there is no reason to make people use multithreaded programs, and multi-threaded often lead to decline in system efficiency. Today, multi-processor systems or "multi-process system" (such as multi-core, hyperthreading, etc.) are increasing, and multithreading has become the future demand. In theory, in multiprocessor systems, multi-threaded programs are much higher than single-line programming. Of course, you must modify a single-thread program to multi-threaded programs and effectively manage these threads, which requires a lot of extra work, which is one of the main things we need to do in the future.
The Drapper Awarder finally believes that new software ideas such as unlimited storage, standardization, multi-threading will change the future of software design, and will also change our future. He finally said to the college students at the seminar: "I even got you, because you have the opportunity to influence and create the future. In the future, this industry will be active as 35 years ago."
(Computer World Daily No. 44 A25)
