How many calculation modes can the crowd circle selection system based on maxcompute + hologres be divided into?

Tag filtering is divided into equivalent filtering , It can be used Equal/In/Between, These filters can be operated at the level of 10 billion . The result set after the operation , We have to do a lot of intersection and combination , Take a common example , A user pays attention to both competing brands and our products , But I didn't buy , In fact, there is no relationship between , Have a bad relationship , Have a close relationship . So the relationship between these groups should be combined , There is a high intersection union calculation . Finally, there is a strong need for accurate weight removal , Because the final calculation result , Become the only one to locate users ID, This ID Delivery that will be used for advertising . So these needs , At the engine level, it is about the efficiency of data reading , If you use line memory to read, will there be IO Enlarge the problem , Data is stored in rows , Real filtering is filtering according to a certain column , however IO Read , Will read the whole line , There will be IO Zoom in on the problem . There will be indexing problems with column storage 、 Filtering effect problem . When calculating the table connection on the operator, it is Hash JOIN The way is still Nest Loop JOIN The way . What is the effect of precise weight removal . These are high requirements for the efficiency of the computing engine . Therefore, the essence is to solve the problem of efficient data storage and filtering 、 Relational computing memory /CPU Consume 、 Accurate de duplication memory /CPU Consumption problem .