who can help
Lao Wang recently bought a VR helmet. He heard that with this high-tech, he can team up with friends to play immersive games, hold face-to-face meetings with work partners, and even experience the work and life of astronauts in space. . But there is one thing, this thing is more machine-consuming, and the server built in my company cannot run. It is said that it is necessary to introduce the concept and technology of cloud computing and cloud rendering into VR business applications, and use the high-speed and stable network to connect the cloud The display output and sound output, etc. are encoded and compressed and then transmitted to the user's terminal device to realize content and rendering on the cloud.
In order to connect my VR helmet to this cloud service, I asked around for help, and finally found out that a guy named "Shuwang Brain" could help.
what is it capable of
This cloud-only VR service is actually a computing power resource. A variety of computing power resources exist in different node positions in the real network, forming what we call a computing power network. The existence significance of this computing power network is to provide users with the most suitable computing power resource service. The most suitable has multiple meanings, including computing power type matching, computing power scale, network connection capability, computing power and network connection overall cost performance etc.
In the early stage, the computing power network experienced two stages of development: cloud-network collaboration and cloud-network integration. With the continuous deepening of digitalization and the widespread implementation of artificial intelligence, the demand for computing power in the whole society has been intensified; with the emergence of edge computing, cloud computing can no longer represent computing power alone, and "cloud" has gradually become "computing". In this way, cloud-network integration has evolved into the current computing-network integration, that is, the computing power network.
Composition of Computing Power Network
Although the computing power network is a combination of computing power and network, Lao Wang is busy looking for network management and cloud management, can he help? The network management only has data on the network, and it can find a way to open a network channel, but it has no idea whether the target computing power resource pool it reaches is suitable; the cloud management only knows the computing power type and capacity in the computing power pool, and it can tell the user Whether this computing power meets the application requirements, but how to get to the computing power resource pool from the user end is a blind road. What's more terrible is that there are multiple independent domains in the network and calculation, and there are network administrators and cloud administrators in different domains in charge, and no one obeys the other.
The brain of the computing network is a master who has both sides. Although it does not directly manage the network and calculation of each domain, it has a very good relationship with everyone. Each network domain and computing power domain will inform the computing network brain of the capability type, available capacity, path and usage status of their respective management domains according to the agreement. The most important thing is that they are very obedient to instructions. How the brain of the computing network allows them to open services, they all construct according to the requirements and never disobey the instructions. The computing network brain has mastered the complete intelligence information in the computing power network, and can command everyone, so it is a simple matter to find a suitable server resource pool for Lao Wang's VR helmet.
Since he has such ability, Lao Wang entrusted this task to the brain of Suanwang.
Choose the right target
After the computing network brain receives the task, it first analyzes the requirements of the task list, including the interface parameters and quantity of the helmet, the type and scale of the VR conference, and computing requirements (such as 8K H.265 real-time hardware decoding, 2K H.264 real-time Hard coding, multi-channel parallel computing capabilities, etc.), and convert these into a list of specific computing power requirements for CPU, GPU, storage, memory, etc. This process is called computing power deconstruction. The location of the deconstructed computing power may be deployed at multiple points, such as the central cloud is responsible for content generation, and the edge cloud is responsible for video decoding and content rendering.
The computing network brain will search through the service resource pool information it has at hand, no matter whether the resource pool is far outside the province or in other cities in the province, or it is in the edge site computer room of the local city. With this stack of checklists, first do an entry qualification screening. The CPU must be 64 cores after the eleventh generation of Intel, the memory only needs to be 256G of DD5, the graphics card must be NVIDIA Tesla GPU, and the storage must be SSD 2T, etc. In addition, for To avoid the complete collapse of the virtual host due to infrastructure failure, the resource redundancy in the resource pool should be at least 100%.
Then the computing network brain will carefully investigate the resource allocation rate (oversale ratio), resource usage rate, resource failure rate, PUE, resource pool operation and maintenance cost of each resource pool, and even whether the power used by the resource pool is green and renewable energy.
For some server-side applications that need to be deployed independently, the computing network brain will also check whether the deployment image files of these applications are pre-stored in these resource pools, and then it can schedule the cloud management to deploy these server-side applications on the newly applied machines. The deployment and commissioning are all completed.
After considering everything in this way, the brain of Suanwang screened out more than a dozen candidates from the hundreds of resource pools at hand, and sorted them according to different evaluation dimensions.
find a good starting point
Once the destination to be selected is found, the computing network brain next considers where to start. In order to use the remote VR computing service, the terminal helmet must be connected to the network. Network requirements can be obtained through scenario analysis, such as bandwidth 100Mbps, RTT 20ms, packet loss rate <0.1%. There are a variety of network access methods, including mobile access and fixed wired access, both economical and fast, and luxurious and reliable. Computing Network Brain will base on the connection capability characteristics of the VR helmet, application scenarios, and Pharaoh’s economic affordability , recommend a suitable solution.
If Lao Wang is a business traveler and the VR helmet supports wireless access, then the computing network brain will find a path for him to access the 5G private network. First, determine which cities the user’s business trips are in, and then evaluate how much bandwidth and low latency the VR game or conference application will have, and then find a bandwidth, latency and latency from the VR dedicated slices in these cities. Slices with sufficient user capacity, and finally register Pharaoh’s user number information. If you encounter a local tyrant who is very concerned about user privacy, then you will first locate the specific geographical location of the user using the private network, and find a 5G base station with good signal coverage, sufficient transmission bandwidth, and stable and normal operation without faults from the surrounding area. Groups, and let these base stations reserve sufficient dedicated PRB wireless access resources for Lao Wang, and use these dedicated resources to build an exclusive 5G slice.
If Lao Wang uses VR headsets in one place in the studio, then the computing network brain will recommend a wired access solution for him. When connecting to the computing power network by cable, it is also necessary to first evaluate the bandwidth requirements and low latency requirements of the VR application, and then find the most suitable transmission access site near the studio. The so-called suitability includes whether the optical path has been pre-covered (which can save a lot of optical path construction time and allow users to enjoy VR computing services as soon as possible), whether the transmission access device has idle ports with matching bandwidth and type, and whether the device has alarms or failures , Whether the transmission equipment has formed a loop (the service will not be interrupted if the network is disconnected in any direction on the loop), etc. If it happens that Lao Wang is a user who requires zero faults and high reliability, the computing brain will find a dual-route access network, and the dual CEs use different optical paths to connect to the transmission equipment on two different loops. No matter how high the user's requirements are, the computing brain will always find a way to find an access solution.
Cross-domain routing calculation
It is said that all roads lead to Rome, not to mention that Rome is not the only goal. From the perspective of the computing network brain, there are actually multiple network domains between the client end and the target computing power resource pool. Although there are many choices, it is not a difficult task for the computing network brain that has mastered the information of each domain.
End-to-end calculation and network selection
The computing network brain first grasps the connection information between various network domains, which two domains can directly communicate with each other, which domains can directly enter the computing power resource pool, and which domains are close and provide intermediate connections. These domains include not only the division of different professions such as transmission and data, but also the division between geographical regions (such as the metropolitan area network of different cities, and the provincial backbone network). The computing network brain accurately grasps the specific connection point information between two adjacent domains, such as how many links are there between which pair of cloud PE devices in the cloud private network and the DCGW device in a certain resource pool, and whether there are different planned loads between them business, or TRUNCK into a logical connection.
The computing network brain will also obtain information in each network domain through the network management of each network domain, including whether there is an accessible path between two points (access point and cross-domain connection point), whether the path has different routes, and Whether the working bandwidth of protection tunnels and tunnel routes meets business requirements, and whether there are faults or alarms in each network node device on the route, etc. For some new network domains, the network management will be turned into a network controller. These controllers can provide a simulation computing tool to the computing network brain, which can simulate the bandwidth, bandwidth, and Estimated values of delay, jitter, and packet loss rate provide more accurate calculation and evaluation capabilities.
In this way, by relying on the two axes of inter-domain network selection and intra-domain routing, the computing network brain has successfully found optional channels from users to multiple target computing power resource pools in a network domain and computing power domain. In order to avoid the entanglement of choosing difficulties, the computing network brain uses a multi-objective algorithm to compare usability, balance, user experience, cost and other aspects, and quickly find a solution for Pharaoh that comprehensively meets all requirements .
end-to-end provisioning
Lao Wang took the solution given by the computing network brain, which computing power resource pools to use, what to use to access the network, and how to span multiple network domains in the middle, and wrote all the details clearly. Lao Wang was very satisfied and decided to place an order for VR services.
Computing Network Brain Intelligent Routing Solution
After receiving the order from Pharaoh, the computing network brain will immediately notify the relevant cloud management to apply for computing resources and deploy VR services on the central cloud or edge cloud according to the requirements of the plan, and notify the business orchestrator or network controller of the relevant network domain , to open the network channel segment by segment.
Commanding and dispatching is not a simple matter. The brain of the computing network must transmit dependencies according to the parameters activated in each domain, and automatically coordinate the activation sequence and parameter transmission. For example, the network in the cloud is activated first, and then the VPN name and VLAN code value are transmitted to the Open the cloud private line channel, and then pass the gateway IP to the terminal to open. If there is a problem in any domain, you have to instruct everyone to roll back what has been done to eliminate the problem of dirty data. Relying on its excellent coordination ability, Suanwang Brain completed all the services ordered in Lao Wang's order within a few minutes as required.
Afterword
Lao Wang just used VR glasses to hold a face-to-face meeting with his work partners, and the whole process was smooth and smooth. But after the meeting, he heard that the meeting server of the company next door was attacked by the network, which made him feel a little worried.
At this time, the mobile phone received a message from the brain of the computing network, "Detecting that the VR service has been accessed abnormally 184 times in the last 24 hours. If you need security reinforcement, please reply 1."