How to get to Guangzhou Baiyun world leather trade center
Publish: 2021-05-17 12:50:20
1. Metro Line 4 Daxing line → Metro Line 7
the whole journey is about 56 minutes / 16.39 km / ticket price is 5 yuan
Wangjing Huacai metro station
988 meters
walk to Haidian Huangzhuang station
Haidian Huangzhuang station last bus 22:20
12 station
(entrance D southwest entrance) take Metro Line 4 Daxing line, Get off at Caishikou station
the last bus at Caishikou station is 22:25
station 1
take Metro Line 7 and get off at Guanganmen inner station (exit C)
Guanganmen inner station
121m
walk to Guanganmen
Guanganmen
the whole journey is about 56 minutes / 16.39 km / ticket price is 5 yuan
Wangjing Huacai metro station
988 meters
walk to Haidian Huangzhuang station
Haidian Huangzhuang station last bus 22:20
12 station
(entrance D southwest entrance) take Metro Line 4 Daxing line, Get off at Caishikou station
the last bus at Caishikou station is 22:25
station 1
take Metro Line 7 and get off at Guanganmen inner station (exit C)
Guanganmen inner station
121m
walk to Guanganmen
Guanganmen
2. Take 974 from Dongmen to the south of Shuangjing bridge, then transfer to 627 to Panjiayuan Bridge. Get off the bus and it's not far from the international eye building
come out from there, take 627 to Nanhu South Road, and then transfer to 606 to Wangjing West Road
tip: before getting on the bus, ask the conctor if you are sitting in the right direction. Don't take the wrong road.
come out from there, take 627 to Nanhu South Road, and then transfer to 606 to Wangjing West Road
tip: before getting on the bus, ask the conctor if you are sitting in the right direction. Don't take the wrong road.
3.
The computer is one of the greatest inventions of our time. It plays an increasingly important role in our proction and life. Now the computer has entered thousands of households and become one of the necessary electrical appliances. Since it is an electrical appliance, there will be some problems. A computer includes a host, a monitor and peripherals. The failure of any part of them will affect the operation of the whole computer
if the host computer of the computer can be started normally, only the display does not respond< first of all, it should be the display failure the monitor is one of the important parts of the computer. If the monitor can't be started normally, it is likely that the monitor is out of order and needs to be overhauled< Secondly, the connection between the host and the monitor is loose if the power indicator of the monitor is not on, it is likely that the connecting line between the host and the monitor is loose or falling off, so it is necessary to check whether the connecting line is connected normally< finally, the monitor is not powered on the monitor also needs power to start, so if the monitor does not respond, it may be that the monitor is not plugged in, so it is necessary to check whether the power plugs of some monitors are connected{ RRRRR}
3. Display power failure< p> The monitor also needs to be plugged in to start, so if the monitor does not respond, there may be something wrong with the power plug of the monitor, which leads to the power failure of the monitor. It is necessary to check whether the power plug of the monitor is powered onWhat do you think is the most likely reason why the monitor doesn't respond
4.
bus line: Metro Line 2 → Metro Line 6 → east section of metro line 14, the whole journey is about 18.4km
1. Take Metro Line 2 from Beijing station, pass 2 stations, reach Chaoyang Gate Station
< P > 2, walk about 130m, transfer to Metro Line 63, take Metro Line 6, pass 3 stations, reach Jintai Road Station
4, walk about 140m, transfer to east section of metro line 14
5 Take the east section of metro line 14, pass 8 stops, reach donghuqu station
6, walk about 280 meters, and arrive at Wangjing huacai holiday...
in Beijing
5. The first step is to re connect the VGA cable and power cable, re plug it, and turn on the monitor switch. If not, try the second step
the second step is to remove the display card, remove the st and visually check the appearance of the display card components. If the capacitance bulges, the display card will burn out. Change the same type of capacitor
the third step is to replace the parts, connect the output of the host computer to the TV, and check
welcome to download Tencent computer manager to learn more and experience more functions.
the second step is to remove the display card, remove the st and visually check the appearance of the display card components. If the capacitance bulges, the display card will burn out. Change the same type of capacitor
the third step is to replace the parts, connect the output of the host computer to the TV, and check
welcome to download Tencent computer manager to learn more and experience more functions.
6. Generally speaking, there are several popular distributed computing frameworks (platforms) in instry (academia):
1: maprec (MR), which is the most popular distributed computing framework
General and its open source implementation Hadoop has been widely used (Facebook,
Yahoo! At the same time, there are many projects developed on the basis of Hadoop (hive is the best one). In addition, there are many companies developed on the basis of Hadoop, such as
cloudera, hortonworks and MAPR
2: pregel, like Mr, is also invented by Google. Its advantage is that it can be more efficient in completing some applications suitable for abstract graph algorithm. Giraph can be regarded as a good open source implementation in development
3: storm, a twitter project, is known as Hadoop's real-time computing platform. For some jobs that need real time performance, it can have higher efficiency than Mr
4: spark, UC Berkeley amplab project, which makes good use of heap in JVM, has better cache support for intermediate calculation results, so its performance is much higher than Mr. Shark is a project similar to hive
5: Dryad and scope,
are Mr (Microsoft Research) projects. From the paper point of view, Dryad is a more general
purpose computing framework, which realizes computing in vertices and communication through channels, forming a graph
workflow; Scope is a bit similar to hive and shark, which compiles some SQL like script
language into jobs that can be calculated on the underlying distributed platform. However, because these two projects are not open source, there is not much information, and there is no community like open source projects
of course, there are many others, such as Google's Dremel and Yale's Hadoop dB (now commercialized as hadapt).
1: maprec (MR), which is the most popular distributed computing framework
General and its open source implementation Hadoop has been widely used (Facebook,
Yahoo! At the same time, there are many projects developed on the basis of Hadoop (hive is the best one). In addition, there are many companies developed on the basis of Hadoop, such as
cloudera, hortonworks and MAPR
2: pregel, like Mr, is also invented by Google. Its advantage is that it can be more efficient in completing some applications suitable for abstract graph algorithm. Giraph can be regarded as a good open source implementation in development
3: storm, a twitter project, is known as Hadoop's real-time computing platform. For some jobs that need real time performance, it can have higher efficiency than Mr
4: spark, UC Berkeley amplab project, which makes good use of heap in JVM, has better cache support for intermediate calculation results, so its performance is much higher than Mr. Shark is a project similar to hive
5: Dryad and scope,
are Mr (Microsoft Research) projects. From the paper point of view, Dryad is a more general
purpose computing framework, which realizes computing in vertices and communication through channels, forming a graph
workflow; Scope is a bit similar to hive and shark, which compiles some SQL like script
language into jobs that can be calculated on the underlying distributed platform. However, because these two projects are not open source, there is not much information, and there is no community like open source projects
of course, there are many others, such as Google's Dremel and Yale's Hadoop dB (now commercialized as hadapt).
7. Distributed computing can be divided into the following categories:
traditional C / S model. Such as HTTP / ftp / SMTP / Pop / DBMS server. The client sends the request to the server, the server processes the request and returns the result to the client. The client is active and the server is passive. This kind of call is explicit. Remote call is remote call, and local call is local call. You have to be clear about every detail. It's not ambiguous at all
cluster technology. In recent years, the computing power of PC is developing rapidly, while the computing power of server is far behind the requirements of client. This kind of many to one relationship is unfair. People have realized that by improving the computing power of a single server, we can always meet the performance requirements. A technology called cluster has emerged, which connects multiple servers and uses them as one server. The advantage of this technology is that it is transparent not only to the client, but also to the server software. The software can run on the cluster without any modification. The application scope of cluster technology is limited to this, which can only improve the computing power of the same software, but can't do anything for multiple different software to work together
general distributed computing environment. For example, CORBA / DCOM / RMI / DBUS and so on, almost all of these technologies (specifications) have network transparency, and the called method may be in another process or on another machine. The caller basically doesn't care whether it's a local call or a remote call. Of course, it is this kind of transparency that leads to the abuse of distributed computing. Distributed computing is easy to use and we think it is free. In fact, the cost of distributed computing is considerable. It is said that the speed of cross process calls may be reced by one order of magnitude, and that of cross machine calls may be reced by two orders of magnitude. Some experts suggest recing the use of distributed computing, even if it is used, coarse-grained calls should be used to rece the number of calls
there are also some mixed forms (soap?), I won't say more here. We mainly introce the third distributed model, which is suitable for both enterprise applications and desktop applications. Some focus on Enterprise Applications (such as CORBA), some focus on desktop environment (such as DBUS). Their implementation principles are almost the same, basically based on the traditional RPC or RPC like implementation, the following describes their basic principles
let's take a look at the simplest model of distribution:
take a look at the traditional request mode of C / S model. The client sends parameters to the server through the network. The server completes the corresponding service according to the parameter requirements, and then returns the results to the client. When the client gets the results, the request is completed at one time. From this point of view, it seems not difficult to call remote objects. The problem is that this method is not transparent to the network. You have to deal with every detail by yourself, which is very complicated
to simplify the design of software, of course, the network operation is transparent, and the caller and implementer do not need to care about the network operation. To achieve this, we can do the following:
introce a proxy object in the client. It has full power to proxy the actual object. The caller doesn't even know that it is a proxy and can call this object just like calling a local object. When the caller calls the proxy function, the proxy does not do the actual operation, but packages these parameters into a network packet, and sends the packet to the server through the network
a stub object is introced into the server. After receiving the data packet sent by the proxy, the stub unpacks the data packet and reorganizes it into a parameter list, and calls the function of the actual object with these parameters. The actual object performs related operations and returns the results to stub, which then packages the results into a network packet and sends the packet to the client's proxy through the network
after the proxy receives the result packet, it unpacks the packet as a return value and returns it to the caller. So far, the whole operation is completed. How about simplification
proxy hides the network operation of client and stub hides the network operation of server, which makes the network transparent. You may say that there is no simplification at all, but the network operation is isolated, and you still have to implement the proxy and stub objects, which is the same trouble
Yes. However, if we carefully study the functions of proxy and stub, we will find that for different objects, these operations are almost the same, just packaging and unpacking, monotonous and repetitive. Monotonous and repetitive things must have rules to follow. If there are rules to follow, you can use code generator to generate code automatically
DCOM and CORBA do the same thing. First, the IDL language is used to describe the object interface, and then the IDL Compiler is used to generate proxy and stub code automatically. The whole process does not need developers to worry about
the technical terms of packing and unpacking are called marshal and unmarshal, which are commonly translated into column set and scatter set in Chinese. However, these two words are too professional to be understood after being translated into Chinese. I think packing and unpacking are more popular
in the above model, the method of calling object does achieve network transparency. Readers may ask, what do I do if I want to access the properties of an object? The attribute of an object is a variable. A variable is a memory area. The memory area is completely independent in different processes. This seems to be a problem indeed. Remember what many books about software design have said: don't expose object properties. If the caller wants to access object properties, he can access them through the get / set method. Can't this be done? Access to properties is converted to calls to object methods
OK, calling object methods and accessing object properties are solved. Another important point is how to create objects. Because the actual object is not fixed on a machine, its position may be dynamic. Even the proxy itself doesn't know where the stub is running. If you want the caller to specify, the process of creating objects has not yet reached network transparency. The usual way is to introce a third-party intermediary, which is fixed and can be found in a certain way. The third-party intermediary is responsible for the thread between the proxy of the client and the stub of the server. There are usually two kinds of third-party Intermediaries: one is only responsible for helping clients find servers, and then clients communicate with servers directly. The other is not only responsible for finding the server, but also for forwarding all requests
the above model is still incomplete, because the real object is not always in a passive position. However, under certain conditions, some events will be triggered and reported to the caller. That is to say, this is a two-way action. The simple C / S model can not meet the requirements, and the P2P mode should be adopted. The original client, as a server, accepts requests from its own server. This is what we do in com. If the client wants to register the event of an object, it needs to implement an IDispatch interface and call the object in reverse
when you implement it yourself, you should consider the following points:
ltransport abstract layer. The distribution may be cross process or cross machine. In different situations, using different communication methods, the performance will be different. It's a good design to make a transport abstraction layer and choose different transport modes in different situations
L text or binary. Package data as text or binary? The advantage of packaging into text is good portability, because people can understand, debugging is convenient. The disadvantage is that the speed is a little slow, and the size of the packed data will be significantly larger. The advantage of using binary is that it is fast, and the data size after packaging is similar to that before packaging. The disadvantage is that it is not easy to debug and the portability is poor
L byte order and byte alignment. If binary transmission is used, portability is a problem. Because the byte order and byte alignment are different on different machines, these instructions should be added to the data package to improve portability.
traditional C / S model. Such as HTTP / ftp / SMTP / Pop / DBMS server. The client sends the request to the server, the server processes the request and returns the result to the client. The client is active and the server is passive. This kind of call is explicit. Remote call is remote call, and local call is local call. You have to be clear about every detail. It's not ambiguous at all
cluster technology. In recent years, the computing power of PC is developing rapidly, while the computing power of server is far behind the requirements of client. This kind of many to one relationship is unfair. People have realized that by improving the computing power of a single server, we can always meet the performance requirements. A technology called cluster has emerged, which connects multiple servers and uses them as one server. The advantage of this technology is that it is transparent not only to the client, but also to the server software. The software can run on the cluster without any modification. The application scope of cluster technology is limited to this, which can only improve the computing power of the same software, but can't do anything for multiple different software to work together
general distributed computing environment. For example, CORBA / DCOM / RMI / DBUS and so on, almost all of these technologies (specifications) have network transparency, and the called method may be in another process or on another machine. The caller basically doesn't care whether it's a local call or a remote call. Of course, it is this kind of transparency that leads to the abuse of distributed computing. Distributed computing is easy to use and we think it is free. In fact, the cost of distributed computing is considerable. It is said that the speed of cross process calls may be reced by one order of magnitude, and that of cross machine calls may be reced by two orders of magnitude. Some experts suggest recing the use of distributed computing, even if it is used, coarse-grained calls should be used to rece the number of calls
there are also some mixed forms (soap?), I won't say more here. We mainly introce the third distributed model, which is suitable for both enterprise applications and desktop applications. Some focus on Enterprise Applications (such as CORBA), some focus on desktop environment (such as DBUS). Their implementation principles are almost the same, basically based on the traditional RPC or RPC like implementation, the following describes their basic principles
let's take a look at the simplest model of distribution:
take a look at the traditional request mode of C / S model. The client sends parameters to the server through the network. The server completes the corresponding service according to the parameter requirements, and then returns the results to the client. When the client gets the results, the request is completed at one time. From this point of view, it seems not difficult to call remote objects. The problem is that this method is not transparent to the network. You have to deal with every detail by yourself, which is very complicated
to simplify the design of software, of course, the network operation is transparent, and the caller and implementer do not need to care about the network operation. To achieve this, we can do the following:
introce a proxy object in the client. It has full power to proxy the actual object. The caller doesn't even know that it is a proxy and can call this object just like calling a local object. When the caller calls the proxy function, the proxy does not do the actual operation, but packages these parameters into a network packet, and sends the packet to the server through the network
a stub object is introced into the server. After receiving the data packet sent by the proxy, the stub unpacks the data packet and reorganizes it into a parameter list, and calls the function of the actual object with these parameters. The actual object performs related operations and returns the results to stub, which then packages the results into a network packet and sends the packet to the client's proxy through the network
after the proxy receives the result packet, it unpacks the packet as a return value and returns it to the caller. So far, the whole operation is completed. How about simplification
proxy hides the network operation of client and stub hides the network operation of server, which makes the network transparent. You may say that there is no simplification at all, but the network operation is isolated, and you still have to implement the proxy and stub objects, which is the same trouble
Yes. However, if we carefully study the functions of proxy and stub, we will find that for different objects, these operations are almost the same, just packaging and unpacking, monotonous and repetitive. Monotonous and repetitive things must have rules to follow. If there are rules to follow, you can use code generator to generate code automatically
DCOM and CORBA do the same thing. First, the IDL language is used to describe the object interface, and then the IDL Compiler is used to generate proxy and stub code automatically. The whole process does not need developers to worry about
the technical terms of packing and unpacking are called marshal and unmarshal, which are commonly translated into column set and scatter set in Chinese. However, these two words are too professional to be understood after being translated into Chinese. I think packing and unpacking are more popular
in the above model, the method of calling object does achieve network transparency. Readers may ask, what do I do if I want to access the properties of an object? The attribute of an object is a variable. A variable is a memory area. The memory area is completely independent in different processes. This seems to be a problem indeed. Remember what many books about software design have said: don't expose object properties. If the caller wants to access object properties, he can access them through the get / set method. Can't this be done? Access to properties is converted to calls to object methods
OK, calling object methods and accessing object properties are solved. Another important point is how to create objects. Because the actual object is not fixed on a machine, its position may be dynamic. Even the proxy itself doesn't know where the stub is running. If you want the caller to specify, the process of creating objects has not yet reached network transparency. The usual way is to introce a third-party intermediary, which is fixed and can be found in a certain way. The third-party intermediary is responsible for the thread between the proxy of the client and the stub of the server. There are usually two kinds of third-party Intermediaries: one is only responsible for helping clients find servers, and then clients communicate with servers directly. The other is not only responsible for finding the server, but also for forwarding all requests
the above model is still incomplete, because the real object is not always in a passive position. However, under certain conditions, some events will be triggered and reported to the caller. That is to say, this is a two-way action. The simple C / S model can not meet the requirements, and the P2P mode should be adopted. The original client, as a server, accepts requests from its own server. This is what we do in com. If the client wants to register the event of an object, it needs to implement an IDispatch interface and call the object in reverse
when you implement it yourself, you should consider the following points:
ltransport abstract layer. The distribution may be cross process or cross machine. In different situations, using different communication methods, the performance will be different. It's a good design to make a transport abstraction layer and choose different transport modes in different situations
L text or binary. Package data as text or binary? The advantage of packaging into text is good portability, because people can understand, debugging is convenient. The disadvantage is that the speed is a little slow, and the size of the packed data will be significantly larger. The advantage of using binary is that it is fast, and the data size after packaging is similar to that before packaging. The disadvantage is that it is not easy to debug and the portability is poor
L byte order and byte alignment. If binary transmission is used, portability is a problem. Because the byte order and byte alignment are different on different machines, these instructions should be added to the data package to improve portability.
8. Speaking of my understanding, my understanding is also relatively simple. Cloud computing means that many hosts provide services at the same time. Because there are many servers providing the same service, it's like there is a cloud behind each service. In distributed computing, e to the limited computing power of a computer, the traditional computing task is divided into several sub tasks by load balancing server, and then given to different computers for calculation, and finally the results are unified. Parallel computing, literally. Follow up: I can understand what you said. I mainly want to ask, as a master's research direction, what is the specific relationship between them. Answer: I feel that it's all around cloud computing. Cloud computing is actually distributed computing. Provide services to many companies and indivials at the same time through distributed servers. This will make full use of resources. I'm sure you know why. I think the emphasis of parallel computing is on coordination. General computing is not only the problem of dividing tasks, but also the problem of time efficiency. Therefore, it is necessary to ensure that all tasks can be completed at almost the same time, so as to prevent this server from completing the calculation and waiting for another one to complete the calculation. Distributed computing plus parallel computing is the real cloud computing. However, when we talk about these calculations now, they are actually equivalent. It's just a superficial personal understanding. If you really want to develop in this direction, it is recommended to go to CSDN more. Now cloud computing is very popular, and every big company is doing it.
Hot content