Btoken virtual currency

8. At this year's yunqi conference in Hangzhou, Dr. Wang Weitao, a geophysicist from the Seismological Bureau of the China Pavilion, made a theme sharing at the data storage technology practical session of Tech insight: the application of cloud computing in seismological research - creating virtual earthquakes using BCS and massive data. He said that the calculation task of five-year data between thousands of seismic stations in China, which originally took one year to calculate, was rampant in cloud computing, and the calculation was completed within 48 hours
how is this achieved
our ancestors watched the stars shine, but it took ten thousand years to figure out the laws of celestial bodies
our predecessors sat and watched the tide level, but it took thousands of generations to sail to the other side of the ocean
however, we are still ignorant of the earth under our feet
it is almost the whole logic of the history of human science from observation and record to law prediction
but every time we try our best to write down the data, it is just a trickle of knowledge
when we begin to calculate, we are looking forward to the moment when Archimedes jumps out of the bathtub and Newton raises the apple
Dr. Wang Weitao is the author of this calculation. He is from the Seismological Bureau of China. What he wants to know is every detail of the earth under our feet< Every earthquake we experience reminds us of the urgency of forecasting and warning this kind of disaster. However, we are still far away from this goal
in order to better understand the physical phenomenon of earthquake, we need extremely detailed crustal structure image, and in order to draw this underground map, we need detailed data calculation. So far, the deepest well drilled by human beings is the Kela ultra deep well drilled by the former Soviet Union, which is about 12.2km. However, the focal depth of earthquakes is often more than ten to dozens of kilometers underground. The current science and technology can not directly observe the focal depth
therefore, we need to rely on thousands of seismic stations distributed throughout the country to detect seismic waves. The propagation characteristics of seismic waves underground are affected by geological structure, which is also the principle that seismic waves can be used to draw underground images. These seismic stations can sense the "big seismic waves" of earthquakes, and also capture the daily "earth noise", such as the vibration of ocean tides hitting the mainland< According to the principle of underground imaging by seismic waves, Wang Weitao told reporters that geophysicists like him are almost half programmers. Because from seismic wave to underground imaging, we have to go through large-scale program calculation beyond the imagination of ordinary people. His calculation model is like this: each earthquake will be transmitted to each seismic station from near to far, so theoretically, each seismic station will make its own record of the same earthquake, and these data are both different and related
using these data, we can calculate some "virtual earthquakes". By comparing and calculating the cross-correlation data between every two seismic stations, we can obtain the valuable data needed to study the overall underground structure in China
virtual earthquake can simulate the same data as real earthquake, so it can be used for underground imaging in areas where there was no earthquake. Each seismic wave data has e, N, Z (east-west, north-south, vertical) components. More than 2000 permanent and temporary seismic stations in China have 6000 components, and the annual data volume is about 30tb, And our total amount of data has reached the Pb level
since we need to compare each component data of each seismic station at each time point, the amount of calculation increases exponentially. Wang Weitao's wisdom and experience are reflected in the programs and algorithms he designed. But Dr. Wang Weitao, who spent a lot of effort to complete the algorithm, found that he had just stepped on the first step of the Long March, and there was still a huge difficulty ahead of him
the connection (gray) between every two seismic stations in the figure is the data to be calculated, and the total amount of calculation is extremely large. If a single computer is used to calculate these data, it will take about seven years. According to a person's career of 20 years, I can only complete three calculations before retirement. In this case, large-scale distributed cloud computing seems to be the only choice< However, the mechanism of cloud computing is not as light as it sounds. The reporter also interviewed the children's shoes of Alibaba cloud, a partner of the China Earthquake Administration. In their eyes, cloud computing, like scientific research, integrates the best wisdom of human beings
required storage space, amount of computation and estimated time required for stand-alone computing
distributed storage: games about farms
cloud storage is like a big farm, each server is like a worker, and your data is sheep. Said Cheng Zong, senior expert of Alibaba cloud storage. Looks like he's a rancher“ Distributed storage "can be regarded as the basic condition of distributed computing. That is to say, your sheep should be put into alicloud's "farm" first, and its workers will help you care for, feed, shear and spin
for Dr. Wang Weitao's data, just storing it in the cloud requires countless "black technologies"
in the future computing, the computing system will access the storage system on a large scale. But these visits must hit the server equally, and there must be no hot spots. But this is not enough, because the server hardware failure in a large-scale cluster will become a normal event, so we must do a good job in real-time resource scheling and fault tolerance
for example, make sure that when one hard disk is removed, the rest of the hard disks should be quickly recovered with backup data
Chengzong cited the above two examples. These two examples can be translated into the analogy of a farm, which is roughly as follows: the workload of the farm workers should be evenly distributed, and the situation of "collecting wool from a sheep" should never happen. In addition, every day, some farm workers fall ill and ask for leave, so their work should be reasonably distributed to many people in the shortest time, so that other workers will not be overloaded
the whole alicloud distributed file system is named Pangu. In Chengzong's mind, Pangu also has many intelligent "black technologies"
he gives an example: we all see the same disks, but Pangu sees different disks. It will score the health of each disk according to the accumulation of historical access data, such as the speed and efficiency of writing
for the disks with poor health, some work assignments should be reced accordingly. These underlying technologies can prepare for Dr. Wang Weitao's next real calculation< According to Cheng Zong, data bandwidth has become an important parameter in distributed computing. From Dr. Wang Weitao's point of view, if you store data on your own server and only use Alibaba cloud's computing power to output results, it can't be achieved. The principle is very simple. All servers of distributed computing send data reading requests to a storage unit, and the bandwidth will be blocked instantly. No matter how powerful the computing power is, it will not be able to play
as for the specific data, the bandwidth of 100 MB optical fiber is generally 100 MB / s, while the bandwidth of hard disk can reach several GB / s at most, and the access bandwidth of Alibaba cloud storage intranet (within the cloud computing system) can reach TB / s level
batch computing: build a pyramid
next, Dr. Wang Weitao's data will enter the final calculation. I am familiar with the Linux system that I am used to. All the computing code is completed in this environment. How to make my code play a role in the cloud computing environment is a very important problem
calculation process of underground imaging data
in seismic science research, alicloud obviously has no way to put forward algorithm suggestions, so what they need to do is to provide a common interface, so that Wang Weitao can use his computer, interface and Linux system in his computer room to control the computing on the cloud
Lin Heshan, a computing expert at Alibaba, is proud of the compatibility and adaptation capabilities provided by Alibaba cloud. Dr. Wang has never used a distributed cluster or "supercomputing" before, so it will be a challenge for him to go directly to the cloud in terms of operation and control
the computing interface we provide allows stand-alone programs to execute efficiently in the cloud environment without modification. With a few simple commands, users can mobilize large-scale computing resources in the cloud for analysis without learning complex distributed computing knowledge. In fact, many people who transition from other places to cloud computing will have such problems, so not only Dr. Wang, but also many other users will use our general computing interface. He said. At this time, the obstacles to large-scale computing are basically cleared
however, Lin Heshan told reporters that the real core technology of cloud computing lies in the computing power scheling of batch computing
the acceleration process and mode of large-scale computing, after the scale of computing is expanded, the access to storage resources will be very frequent. At this time, the control of the concurrency of access should be very "careful"
Dr. Wang's application has many small I / O requests. If every I / O request directly accesses cloud storage, the resulting delay will affect the computing efficiency. In order to further optimize the computing performance, batch computing adopts the "distributed cache" strategy, which caches the data that may be used around the computing nodes in advance. In this way, the computing power is not limited by the cluster size. Lin Heshan said
even so, it is far from enough. It is an important moment to test the intelligence of the system to determine the "granularity" of data access. If one read is too much, it may cause bandwidth congestion. If one read is too little, it will cause frequent access. For different types of data, it is an arous task to make reasonable prediction and read automatically
for example, it's like building a pyramid. Tens of thousands of "slaves" have to work together. The engineer has to decide whether to transport multiple stones to the site at the expense of speed at one time or to transport one stone to the site at a time at the expense of quantity
similarly, in the face of the vast pyramid project, how many slaves should be assigned to mix mortar, carry stones, and build each block? Even engineers need to carefully consider how many slaves can be assigned to the system to complete the task automatically, and the difficulty can be imagined
of course, errors often occur in such a complicated calculation process
Lin Heshan gives an example: when rendering the animation "little door god", alicloud's fault tolerance mechanism plays a role At that time, 2000 servers participated in large-scale batch computing at the peak.) In general, video rendering is a continuous and long process. If there is only one problem in a frame rendering, it will cause large-scale access delay, logical congestion and "hot spots"
Lin Heshan said: "alicloud's approach is to run again in the shortest time after a calculation error, if

9. Tigerwit trading platform is currently in-depth exploration and application. Tigerwit is currently comprehensively exploring and promoting the in-depth application of blockchain technology in the field of foreign exchange trading services, hoping to build a more fair and efficient trading platform with the help of the characteristics of blockchain such as "decentralization", "information distributed storage", "whole network accounting" and "difficult to be tampered with" Transparent foreign exchange trading service mode.