Does the data center require an operating system?

The PC has an operating system. Even mobile phones have an operating system. Why doesn't a data center have an operating system?

Of course, it is much more difficult to make an operating system that can handle all the resources of a data center than to create an operating system that allows a single device to run applications. The data center has a team of IT professionals to ensure that all servers are running and that the application has enough storage and so on. However, the workload is so great that it is necessary to introduce a more expensive operating system that can handle the entire datacenter transaction.

This is the view of the University of California, Berkeley Ph. D., Matei Zaharia, at the USENIX annual Technical Meeting held in Oregon State Portland this week.

He is not the first person to propose an operating system for computing systems clustering. However, he sees this need becoming more and more urgent, as applications and users, programming frameworks, and storage systems are becoming more diverse.

The data center operating system can centralize all of these in one Management platform and provide resource sharing, data sharing, programming abstraction, and debugging.

Zaharia says we develop time sharing and computer operating systems for the same reason.

An audience member pointed out that the idea of manufacturing an operating system for a computer cluster has been in for decades. He dismissed the idea that Zaharia called it a new idea and why it would now win.

Zaharia that earlier versions of the data center operating system were already in production. He points out that Google is already running its data center with advanced technology. The technology has not yet been fully disclosed to the public. Google's software stack is like a design based on an operating system.

Zaharia and his colleagues explained their ideas in a paper titled "Data Centers need an operating system." You can see this paper on the USENIX website.

The paper says that data centers already host many kinds of applications (storage systems, network applications, long-running services, and batch analytics), and as the new cluster programming framework is developed, we expect the number of applications to grow. For example, Google has enhanced its mapreduce framework with Pregel (a special framework for image applications), Dremel (a low latency system for interactive data mining), and percolator (an incremental indexing system). At the same time, the number of computer cluster users is increasing. For example, http://www.aliyun.com/zixun/aggregation/1560.html ">facebook's Hadoop database handles almost simultaneous SQL queries of hundreds of users." It is therefore important for data center operators to be able to efficiently and repeatedly utilize resources between users of an application and multiple applications.

Zaharia did not say they had built a data center operating system. But, he says, his team has taken initial steps to design a cluster manager called "Mesos", which enables accurate sharing between applications.

Some questions that still need to be answered include how to make standardized programs, how to handle streaming data, and how to secure storage performance.

But Zaharia believes many companies, including Google, Amazon and Microsoft, are addressing these issues.

Zaharia in the paper, software platforms such as Hadoop Stack, LAMP, Amazon Web Services, Windows Azure and Google Gfs/bigtable/mapreduce are the de facto data center operating systems. These platforms will evolve to address the diversity of data center users and workloads. However, data center applications are still difficult to develop and are not easily compatible with each other.

(Responsible editor: admin)