But the data center to prepare for the future: hardware transformation will take place in memory (dynamic random access memory), and memory as the data center main memory has a history of more than 20 years.
Looking forward to the future, have a variety of options can improve the efficiency, persistence, and reduce the cost.
In order to solve some problems in the innovation, industry media Search DataCenter with dell technology global vice President of technology strategy, Danny Cobb is discussed.
Cobb, over the years have seen a lot of industry changes, he was once EMC company's chief technology officer, as digital equipment corporation, senior technical personnel, Cobb, summarizes the data center in the future plans for IT professionals focus on all kinds of infrastructure technology.
You has publicly discussed the data center single-stage units to multistage evolution of memory.
What kind of technical data center is crucial for the future?
Danny Cobb: now there is a kind of using artificial intelligence (AI) and the concept of machine learning technology of real-time optimization of infrastructure.
We are actively involved in the graphics processing unit, tensor processing unit, the field programmable gate array (FPGA) and so on some new calculation model for research work, as a fundamentally in architectural services.
Users can use the machine learning and artificial intelligence technology according to the available resources in data center scheduling workload.
Three or four years ago, every workload in homogeneous virtual run on x86, and this is a kind of homogeneous computing world.
This new world is heterogeneous computing, unload the engine, is to accelerate, artificial intelligence is dynamic programming in the FPGA in the data center.
Data center infrastructure itself must therefore carry more and more knowledge, we see the style of this infrastructure as on these platforms and workload of calculation of the style of the development and continuous development.
Decomposition and combination of the infrastructure seems to local deployment for cloud computing data center back to the application, then what is the future of the data center?
Danny Cobb: as an IT professional, the idea is to make more work, get the most value, and in unit time and unit cost of handling most of the data.
Integration of infrastructure to solve the first question is, can now buy a set of IT equipment...
I can predict their performance, but also understand the cost, and enterprise team does not need to do it for me.
Now, I think will be deployed in more fine-grained, these things are more likely to consume the capacity of the block.
It makes us to achieve the integration.
Now, can buy a smaller unit - a 1 u server, you can get some management and the ability of the business process, so that the hardware management, and deployment of a Shared storage software stack, and a single, unified the longitudinal extension of the storage space.
Today, both Intel and AMD or other architecture, fundamentally speaking, people through the DDR (double data rate) tightly coupled memory for processing.
However, there are some in the industry and the technology roadmap such examples, such as GenZ, OpenCAPI and C6 bus technology.
This is we have already begun to traditional memory hierarchy and the separation of a processing model, this model can realize flexibility.
Technology such as PCIe (PCI Express) is basically I/O bus technology, you will realize the bandwidth double every two years, and to reduce half of the delay.
This is a great single bus system.
And more bus system is not really a structure, it does not have the ability to configure itself, and cannot use the other fiber optic technology equipment in real time.
Space, the new bus that is RDMA Ethernet and use it as a new system structure of ability.
This is also I mentioned before the memory bus, C6 and GenZ.
沒有留言:
張貼留言