Video tour in the data center: The strategy of creating data centers

Intel has come a long way before developing a data center strategy. Historically, it was that in those places where the company bought or built buildings, data centers were also located for the needs of the development departments there. Experience has shown that it is not very effective. With this approach, the number of Intel data centers has exceeded 150 and, accordingly, they have become a very expensive part of the corporation’s resources. Therefore, a number of decisions were made to reduce the total cost of ownership of data centers. Let's take a closer look at a number of techniques that have resulted in the formation of a common strategy for Intel's data center.

1. Timely proactive server server replacement

The company has a four-year server update cycle. This technique has proven its effectiveness. Depending on the generations of processors, the ratio of the number of new and replaced servers changes, sometimes for example 1 to 10 or 1 to 15. As a result of the upgrade, on the one hand, serious electricity savings are achieved as new systems are more energy efficient, on the other hand, productivity increases.

2. Resource Virtualization

This is especially true of the part of the equipment that provides office activities. A simple example - there is a Web server, mail server, DNS server. Each of them loads the resources of a dedicated physical server, say, by 10%. This is inefficient, of course, and if they are placed on one computer through virtualization, it will be cheaper and easier to maintain. Virtualization is the first major step towards GRID computing and, ultimately, the introduction of cloud computing. Intel is very active in building its own internal clouds.
')

3. Server consolidation

Over the past 5 years, the number of data centers has been significantly reduced: from 150 to 93. In addition, the company seeks to reduce the number of server rooms. It is clear that several small server rooms with a load of about 40% in maintenance cost more than one large room with a load of 60-70%. This stage took place at Intel several years ago; a large program of measures aimed at consolidation was implemented, which turned out to be very successful, which made the maintenance of data centers less costly.

4. Increase server utilization

That is, increasing the degree of loading equipment. First of all, attention is drawn to a number of points. Thus, the entire server park of the company is divided into 4 parts according to the directions of use - Design, Office, Manufacturing, Enterprise (abbreviated DOME). The Manufacturing segment is practically not involved in this process, since it provides support for the work of the factories. As for the other classes of servers, as a rule, Office and Enterprise are combined together, because they are very similar in tasks and equipment, and Design resources exist separately, that is, design support and development.

Increased utilization of the servers themselves is achieved primarily through the introduction of virtualization. Intel has two main programs for virtualization - DCU (design software virtualization, large server pools) and DCV (virtualization in the Office and Enterprise segments). The greatest interest for other companies, apparently, is the experience of Intel in virtualization program DCV. Here we are talking about the usual practice, when from 4 to 8 virtual machines are put on one physical machine, and the utilization of its resources increases. The efficiency of using such servers is growing, the money spent is being spent faster.

5. The use of effective means of cooling and energy saving

During the construction of new rooms and data centers, as well as during the modernization of existing ones, the energy efficiency of new technologies is of great importance. For example, Intel maintains air cooling in its data centers and there are a number of techniques to reduce operating costs with this approach. We talked about this in four previous articles, we recall - this is the introduction of racks with exhaust hood, free cooling (free cooling), heat recovery system ... Even such basic things as using caps in racks reduces the load on the cooling systems and reduces energy costs.

This strategy is not a clearly defined action plan, but simply the experience of Intel, the result of trial and error accumulated over the years. If at the very beginning the global goal was to reduce the number of data centers to the maximum, almost to 18-20 pieces, then over time it became clear that this is impossible physically and still some amount of equipment should be located in remote places, and not only on large hubs. Because WAN channels at the moment do not allow to consolidate everything to such a high degree, and a lot of things prevent or make impractical such a concentration of server resources. Therefore, it was decided instead of such a global consolidation to move to an increase in utilization of data centers. After all, the main goal is to minimize capital investments in data centers, that is, not to build new ones, but to effectively use existing ones. Building a new data center is a huge amount of money, millions and millions of dollars, depending on its size. And here came a series of techniques for how to achieve the best use of existing resources, including the server update cycle, virtualization, clouds, etc., which has developed into the current strategy for Intel's data centers.

You can ask Leonid Shishlov a question in the community of Intel professionals .