On this side of the ocean, Cai Siqiang was busy with the acquisition, but Duke didn't think about it at all, and now in addition to coping with the course every day, he is thinking about how to get the nuclear reaction simulation program in the nuclear laboratory.

This is so difficult that so far Duke has been able to use the nuclear reaction simulation program through a desktop application, which is the most valuable asset of the MIT Nuclear Laboratory, and its models are based on the data obtained from previous nuclear tests conducted by the Nuclear Laboratory and other nuclear reaction research institutions, most of which have been verified by subcritical experiments in the laboratory.

It can be said that this application is the brainchild of countless scientists in American universities and official nuclear physics research. Although Duke doesn't know if there is a similar system in China, even if there is, I believe that there is a big gap compared with this system, after all, the research foundation of the two sides is different. Whether it is computer modeling or experimental verification, the gap between the two sides is not a star and a half.

And even if there is such a simulation system in China, Duke will never have any chance of contact, and a system like this basically belongs to the category of top secret, unless Duke also decides to be an invisible person.

So, MIT's nuclear reaction simulation program is Duke's only hope.

It's just that this hope seems a little slim.

Duke couldn't find a way to break through the nuclear reaction simulation application system, and Kerry's side has already seen the effect of streamlining the new operating system, which is based on Craylinux, and has successfully run on a simulated two-socket, 20-core workstation platform.

Compared with the original Linux system of this workstation, the new system of Kerry's streamlining and optimization has fully doubled the performance, and what is especially rare is that it retains 100% compatibility with the original Linux system on this basis, which means that the original application can run directly on the new system without any changes.

The new system was reduced to half the size of the original Craylinux, and the resulting binary code was reduced by almost 40 percent.

In the following further simulation tests, the 4-channel and 8-channel simulation systems were also verified, proving that Kerry's streamlined system has fully realized the new layered architecture, at least functionally.

Kerry then extended the system further to other CPU platforms to see if the new architecture worked just as well on other CPU platforms.

Seeing Kerry's remarkable achievements in the improvement of the operating system, Duke suddenly moved in his heart, and instructed Kerry to optimize and improve the operating system of the Titan II prototype to see if he could achieve some results, because Kerry was already very familiar with this platform, and the results were soon available, and after the optimization and improvement of the Titan II prototype, the new Craylinux system was improved by nearly 50%.

This optimization effect scared Duke, this is a supercomputer, so much improvement, it is equivalent to upgrading the system again without spending money. Duke called up the improvements in Kerry optimization and found that the biggest improvement came from the improvement of the parallel scheduling algorithm, and the modification of thousands of lines of code alone directly improved the performance of the system by about 15 percent.

Kerry's improved new parallel scheduling algorithm, using some prediction functions, makes the computing power of the entire supercomputer system almost 100 percent saturated, while the original algorithm can use more than 80 percent of the system at best.

With this result, Duke pondered for an afternoon, and finally came up with a roundabout method, deciding to try his luck with this result to see if there was a chance to get his hands on what he wanted.

"Professor," Duke knocked on Professor Andrew's door, respectfully speaking, "I studied the source code of our supercomputer operating system and found that there is still a lot of room for performance improvement. ”

"Really?" Professor Andrew looked with interest at his protégé, known as a computer genius, and asked, "Is there a way to improve?" Andrew didn't expect Duke to find a solution to the problem so quickly, it was just a lip service.

"I rewrote the code for that weak part, and after simulation testing and calculations, the performance improved by about 15 percent." Duke said.

A fifteen percent increase in performance?! Andrew's originally relaxed face was full of disbelief, and he didn't expect Duke to not only find the problem, but also solve the problem, improving the performance of the system by fifteen percent, you must know that this is a supercomputer, and an increase of one percentage point is extremely difficult, Duke actually said that it can improve the performance by fifteen percent!

Andrew couldn't sit still, he stood up and asked excitedly, "Duke, this matter can't be believed, what evidence do you have?" ”

Duke took out his laptop and pulled up the source code to modify the Craylinux parallel scheduler, "Professor, you see, this is the part of the code, it was originally written like this, according to my calculation of this algorithm, the utilization efficiency of this algorithm for supercomputer computing units should not exceed 85 percent." ”

"However, after this modification, the utilization rate of supercomputer computing units will be increased to 99.0 percent, and the efficiency of scheduling will be greatly improved through intelligent prediction of the use of computing units."

Duke installed a virtual machine software on his laptop, this virtual machine was developed by Cray, which can run Craylinux for debugging, Duke showed Andrew how the system would run before and after the modification, and used a test software to test the performance of the system, and found that the performance of the virtual machine software had increased by nearly 10 percent.

This is because the environment in which the virtual machine runs is a bit different from the real supercomputer, and due to the lack of tens of thousands of computing units, the impact of the improvement of the scheduling algorithm on the performance is not fully reflected, but for Andrew, this is enough.

Being able to increase 10 percent is also a huge step forward in a supercomputer, knowing that the performance improvement of a supercomputer is not an easy thing, behind every percentage increase means paying a high price, if there is now an opportunity to improve the performance of 10 percent without spending a penny of hardware changes, Andrew is really happy to open a bottle of old red wine to celebrate.

"Duke, can you stay with this modified program? We'd like to get in touch with Cray to see if we can use your improved operating system to improve Titan II. Andrew asked. After all, this improvement is a result of Duke, and it is a very significant achievement, and Andrew needs to get Duke's permission to get the source code, although Andrew is still Duke's mentor.

Duke was a little pained about taking out this improvement, after all, Cray could use it on his own new model in the future. Given the high cost of a supercomputer, the new algorithm, which has been improved by as much as 15 percent, is worth hundreds of millions. But now it has to be taken out, because Duke hopes to use this scheduling algorithm to improve and hoist the wolf of the nuclear reaction simulation program.

"Of course, there is no problem, but I hope that this improved intellectual property will stay in the nuclear laboratory. For Titan II only. Duke asked that the use of the improvement be limited to a certain scope, which of course Professor Andrew would be happy to see.