from 2012 Pre research started in ,2014 Build software and hardware collaborative system in ,2017 Applied to huaweiyun in ...
Diachronic 8 Huawei cloud sky architecture , How strong is it , Maybe we can find the answer from the following .
At the end of GECCO 2020 International conferences , Huawei cloud Optimus architecture scheduling algorithm team obtained OCP（Optimal Camera
Placement Optimal camera layout ） and USCP（Unicost Set Covering Problem Single cost set covering problem ） Operational research optimization algorithm track two sub track champion .
Top level meeting beyond imagination
GECCO The meeting started at 1999 year , It is one of the most important events in the field of evolutionary computing . This year's competition attracted people from the UK , France and other world famous research institutions and top scholars , Such as the French optimization solution provider Artelys（ Flagship competition in the field of industrial optimization ROADEF/EURO
Challenge 2018 Champion of the year ）, Lancaster University （ROADEF/EURO Challenge
2016 Champion of the year ）, University of Grenoble, France , University College London, UK, etc , It can be described as a gathering of experts .
Got a general idea of the meeting , Let's take a look at the problems that need to be solved in the winning track of this race .
In Computer Science “ Evolutionary computing ”, Refers to a series of “ Global optimization algorithm inspired by biological evolution ”, And the artificial intelligence of this kind of algorithm , It is mainly used in solving optimization problems . and OCP And USCP As a classical discrete optimization problem , It's proven NP-Hard problem , among USCP what is more Karp Proposed 21 individual NP-Complete One of the problems , It is of great significance in the study of computational complexity theory , And it is widely used in edge site selection , Software fuzzy testing and other practical industrial scenarios .
OCP The problem can be described simply as ： Suppose a city needs to deploy a group of cameras to monitor the full coverage , And where each camera is deployed （400 10000 optional locations ）, The angle and coverage of the monitoring area are not the same , How to use the least number of cameras to achieve full coverage of urban monitoring .USCP The problem is described by a more abstract mathematical set , They are essentially the same .
OCP Problem diagram
Cloud practice and algorithm theory collision
This competition provides the data collection based on the actual urban monitoring layout transformation , The largest of these data contains 380 More than 10000 monitoring candidates . From 380 The optimal layout scheme is selected from ten thousand candidate positions , Search space up to 2^(3,800,000)≈〖10〗^(1,143,913), That number is far more than the total number of atoms in the universe , Even with the world's computing power , It is also impossible to verify the advantages and disadvantages of each scheme in a limited time .
Huge search space , Greatly increase the difficulty of competition questions
Submitted by Huawei cloud Optimus architecture scheduling algorithm team Weighting-Based Parallel Local
Search（WPLS） The algorithm combines the skills of machine learning and operational research optimization , The tabu table strategy is used in the local search process , And self-learning to adjust the evaluation function to jump out of local optimum . In terms of implementation , The algorithm takes advantage of the unique hardware advantages of Huawei cloud Kunpeng and shengteng , It not only brings into play the parallel acceleration ability of the algorithm , It is also found that the scheme is close to the theoretical optimal solution in a very short time .
For how to 380 The problem of selecting the best layout from ten thousand candidate positions , The core problem is how to select a large number of edge sites , Planning the capacity of each site , The global service access experience is optimized through intelligent global scheduling
, Its essence can also be abstracted as a series of optimization problems with the core of set covering problem . Huawei cloud team proposed “ Cloud site location ” The plan of . Plans to deploy massive sites nationwide , Calculate the cause delay , Service quality , The problem of limited coverage area caused by constraints such as actual environment , Calculate the coverage area of each station and the corresponding construction cost , Finally, the optimal site deployment scheme which can achieve full coverage is proposed .
Have to say , The solution benefits from the current cloud technology has become an important thrust of the development of the times . With the development of industry , Industry's influence on massive computing power , The extreme delay experience has put forward higher requirements . Cloud computing as the core production tool in the era of digital economy , Is gradually extending to the edge , In order to meet the surging computing power at any time , Anywhere , Obtain and realize the service access nearby on demand .
Facing the future Full stack technology investment of Huawei cloud sky architecture
Go through 8 Huawei cloud sky architecture with technology accumulation , Through the minimalist data center , Dedicated hardware , Virtualization , Cloud operating system and other full stack technology investment , Provide hard core performance , Ultimate Durability , Excellent performance , Cloud services with cloud edge collaboration , For China, for clouds , Hua Weiyun Stack, Huawei cloud edge provides consistent experience and consistent ecology .
“ Wisdom cloud brain ” As the management and control side of Huawei cloud sky architecture , It's for the cloud ,AI,5G Distributed cloud operating system of the times , To achieve the optimal supply of global resources , Simple use of diversity computing power . among , Global resource scheduling capability can support the future 10 Complex scheduling coordination among ten thousand level distributed stations , Complete center and edge , Intelligent on demand scheduling between edges , Match the optimal nodes according to business demands , Access nearby . For tenants , Smart cloud brain realizes intelligent recommendation of computing power through resource portrait and prediction algorithm , Let the application load run on the most appropriate computing power .
future , Huawei cloud will continue to play its full stack technology innovation capability , Let's work together with our partners , Help government and enterprises realize digital transformation and intelligent upgrading .