Tech

Knowledge post: What is the operation of stacking N PCs into a PC farm?

  ●Personal computer, PC, everyone is familiar with this, office, audio-visual entertainment, play games, do some simple creative work (such as video editing, simple 3D modeling, simple product design), completely OK! The characteristics of PC, from Intel’s side, are mainly platforms equipped with (Core) Core processors.
  ●The next level up is the workstation. For professional design tasks with heavier loads, such as designing a motorcycle engine, a conventional PC can be a bit difficult, so it has to be a workstation! Starting from the level of workstations, Xeon processors have been used. The number of cores is more than that of PCs, and the motherboards used are also different. ECC error-checking memory can be used to avoid downtime during work. Moreover, starting Equipped with a professional graphics card.
  ●What about going up? That is the server, and the server cluster. It uses several or massive Xeon processors, uses Win Server or Linux system, and then according to different application requirements, it will be filled with memory (such as several thousand GB) or hard disk (such as 10PB, 100PB, 1PB=1024TB) , Of course, it can also be filled with accelerator cards for deep learning. In short, complex tasks, such as the rendering of movies and animated feature films, such as large-scale mechanical design, such as game servers, such as data centers, and deep learning training, etc., need to use servers or server clusters!
  But in fact, there is still a relatively large span between workstations and servers (clusters). In practical applications, there is often such a situation: the computing power requirements of certain application types are much higher than that of workstations, but if they are used on servers (clusters), the cost is too high – after all, Xeon processors and their dedicated motherboards Peripherals such as memory and memory are relatively more expensive, and the WindowsServer system is still charged, and many applications on the Server are often charged (even charged according to the number of nodes)-this is quite specific! !
  In order to solve the above-mentioned practical problems, in 2018, Intel created a new technology and new thing: personal computers (PCs) are combined and stacked through racks to gather computing power (and distribute), which can be understood as ” A server built with a PC”. And Intel also gave it a particularly appropriate name called “PC Farm”, which means “PC Farm”.

  This is the “official definition” of PC Farm.

  PC Farm has also undergone many evolutions and iterations, but its unified feature is: it is composed of a complete “PC system”, that is, a personal computer system! In the picture above, you can clearly see three sets of complete personal computer systems: three motherboards, each with two memory sticks, an M.2 interface SSD, and a processor installed, and each There is also a complete set of interfaces on each motherboard.
  Later, PC Farm evolved into a vertical plug-and-plug form to facilitate expansion at any time and more stacking (stacking on the rack to form a cabinet one by one), but it can still be seen that it is composed of a single PC unit.

  The newer PC Farm products are very easy to plug and unplug, but a single component unit is still a complete PC. Of course, some server technologies are also used, such as server cooling technology.
What are the specific advantages of PC Farm?

  ●First of all, the total computing power can be very high! Because it can also be stacked with several cabinets, it can be managed and deployed internally through IPMI (Intelligent Platform Management Interface) or third-party management software based on IPMI to realize the combination (and distribution) of hardware performance, so it The “overall performance” of the server can be very high (144 computing power nodes can be deployed in one cabinet), and it can handle a large number of professional applications.
  ●Secondly, the cost is much lower than that of the server system. This “relatively low cost” has many aspects. First, the PC hardware system is relatively cheap, and the system and software costs are also low. In addition, these PCs used for combined stacking are “complete individuals”, they can be easily replaced, and now they can even be “stacked across generations” (for example, most of the 12th-generation Core platforms before can now be partially upgraded to 13th Generation Core platform, different generations can be mixed). The most exaggerated thing is that the individual that is replaced (eliminated) can be sold as an independent PC at a discount—because its accessories and functions are complete. In other words, the expansion and upgrade of PCFarm is simple and the cost is much lower!
  ● Flexible and fast deployment and management. Because PC Farm introduces the concept of server plug-in expansion and remote management, its deployment and management are relatively flexible and fast. After training, it is actually very easy to use.
  Based on the above characteristics, PC Farm can actually realize many applications. At present, it is mainly active in e-sports Internet cafes, e-sports hotels, cloud games, cloud rendering, VR entertainment and other fields.
  Don’t be fooled by the word “cloud”, it’s actually not that complicated. Uncle Niu will give you a few different types of examples below, I believe everyone can understand: ●E-sports
  hotel. We assume an e-sports hotel with a total of 100 rooms distributed over 5 floors. The most traditional approach is: put a high-performance PC in each room, such as the 12th generation Core i7+RTX 3070/4060. The total amount is 100 units, not to mention the huge investment, but also the risk of accessories being stolen or damaged.
  With the PC Farm solution, we only need to deploy a rack-mounted PCFarm “server” in one room, perhaps using only 30 nodes (30 i7 processors + 30 RTX independent display systems) for unified management and dynamic performance allocation. It can meet the needs of 100 rooms (after all, most users only play online games and do not need much performance). Similar to this situation are Internet cafes.
  ● Another example is a small film and television studio. This is easy to understand. These small film and television studios mainly do special effects videos and complete the editing of massive videos. They have high requirements for processor performance and graphics card performance, and a single PC is definitely not enough. In the past, such studios often bought second-hand servers to use processors for rendering (on a brand new Xeon system, whether it is a workstation or a server, the cost is very high).
  ● Mix and match PC Farm and server clusters to achieve “cloud gaming”.
  This example is a bit more professional, but it is also easy to understand: Suppose a game manufacturer has developed a game with a very beautiful screen (usually an online game), but the problem is that if all calculations are completed on the client side (such as mobile phones, laptops), The requirements for the client hardware are very high, and the client program will be very large, just like today’s large-scale 3A games, with a capacity of hundreds of GB at every turn—this is a nightmare for mobile phones!
  So how to solve it? A better way is to complete the rendering of the game in the cloud, which greatly reduces the computing and storage pressure on the client. However, the job of rendering is not suitable for servers in the traditional sense-the server can use a processor to extend TB-level memory + massive hard disks, but the job of rendering depends on the number of processor cores and the number of GPUs. And performance-this depends on the server system to pile up, but it is expensive! Then the reasonable solution is: the traditional server completes the data part, and the rendering part is handed over to PCFarm to complete!

  This is the “cloud game model”. In fact, the most traditional “web game” is also a typical cloud game. However, in today’s cloud games, the picture quality has been greatly improved-the test is the computing power of PC Farm.
  Xishanju’s “Jian Wang 3” and “Jian Wang 3 Origin” cooperated with Intel and Ali Edge Cloud ENS, relying on the PC Farm cluster to complete the screen rendering, thereby greatly reducing the demand for client hardware.

  Edge computing power, such as surveying and mapping modeling.
  Don’t be intimidated by the “standard industry vocabulary” of “edge”, wait for Uncle Niu to give you a clear interpretation:
  Before 2020, everyone knows an interesting thing called “concert escape”. Many fugitives couldn’t stand the loneliness and went to the singer’s concert, but they were caught on the spot! Not to mention, many fugitives were caught at the concert.
  How did you do it? In fact, the mechanism is very simple: a camera is installed at each entrance and exit of the venue, and the camera captures the faces of the audience entering the venue and compares them with the fugitive database of the Ministry of Public Security. In this system, the camera is the “terminal”, and the fugitive system of the Ministry of Public Security is on the server of the Ministry of Public Security, which is the “cloud”.
  But this method has a real problem: if all faces have to be uploaded to the server of the Ministry of Public Security through the Internet for comparison and then fed back, the speed will be too slow to be released in time. result. Therefore, this system introduces “edge computing power”-each camera or every few cameras is equipped with a computer. This computer not only records the pictures taken by the camera for subsequent comparison in the database of the Ministry of Public Security, but also With the facial feature data of hundreds of thousands of fugitives, it can then complete a quick comparison and screening, which is much faster and more efficient than doing it all through the cloud. In this model, the camera is the “terminal device”; the computer close to the camera is the “edge computing power” (“edge” means that its physical location is closer to the terminal than the cloud); and the database of the Ministry of Public Security is the “cloud computing power”. “.
  However, in the above example, face recognition only captures key features, so the requirements for “edge computing power” are not high, and a high-performance PC is enough. But what if the collected information that needs to be preprocessed at the edge is more complex and requires a higher computing power? For example, surveying and mapping data is not only a complex high-precision image, but also contains a lot of location information such as coordinates and altitude, and it also needs to be identified and analyzed. At this time, the edge end must rely on high PCs to stack performance—PCFarm here It just comes in handy – use a multi-node PC Farm cluster system to cover a range of surveying and mapping information for preprocessing, and finally upload it to the cloud.
PC Farm (cluster), broad application prospects

  Seeing this, I believe that everyone has a comprehensive understanding of the product PC Farm, and what I want to say is: this magical product actually has a very broad application prospect. Someone said before that it is “a product between a workstation and a server”, yes, but it is not true – after all, it has many characteristics that even traditional servers do not have, and its computing power can be quite powerful in theory. Compared with the server system, the total cost of ownership is significantly lower, and it can really show its strengths in many applications-in the end, it depends on how you use your imagination!

error: Content is protected !!