MENGEVALUASI ARSITEKTUR DATA CENTER TERDISTRIBUSI

EVALUATING THE DISTRIBUTED DATA CENTER ARCHITECTURE

Published on 19th December 2014 by Vinay Rathore

Share on linkedinShare on facebook

Share on twitterShare on emailShare on print

I was recently visiting Hong Kong, and as I was going through an industrial district in Kowloon I noticed the construction of a few enormous non-descript buildings, with very few windows. It got me wondering was this some sort of penitentiary or secretive government building? Then I noticed that there was not enough barbed wire to be a jail nor any visual indication to be a secretive government facility. Fortunately, a quick search on cell phone GPS maps advised that it was just another large data center.

While there has been a strong trend toward creating bigger and more powerful centralized data centers to meet end user needs, there has been another, less notable, trend toward pushing the data center closer to end users for the same reason. Large data centers create efficiency through a centralization of resources (e.g., uninterruptable power supplies, generators, HVAC, other networks, etc.). The goal is to minimize cost while maintaining maximum functionality and reliability of the data center itself. However, many end users find that they prefer certain mission critical elements of their IT infrastructure to be physically closer to their offices, rather than in a central data center that could be far away. This has given rise to the need for a data center solution physically closer to the customer (aka virtualized data center), which essentially comprises a network extension of the centralized data center into space that is physically closer to the end user, while still offering many of the traditional data center services. For some locations, such as a remote business park, building a full service, but smaller scale data center may be justified; for others a virtual data center is the next best option.
 

http://pro.clubic.com/entreprises/google/actualite-517039-google-data-center.html

Why large centralized data centers are not for everyone

Today, enterprises are more dependent upon their IT infrastructure than ever before. Further, they no longer want the burden of managing IT complexity, a function typically far from their core competency. Instead, they simply prefer to move IT resources into a location where space, power and network access is abundant and can be managed remotely. For some enterprises, this model creates a dilemma. The question that arises is, “How should I treat my mission critical applications in a data center that is 80 kilometers away, is shared with others, and in some cases is not easily accessible (physically, that is)?” Some business applications don’t care about such parameters, like certain Human Resource functions, or Customer/Marketing applications, but other applications, such as proprietary algorithmic trading, 3D modelling applications, or high volume transaction-oriented processes, may suffer due to latency and need for greater control. The alternative is to keep an IT facility onsite, but ideally there is a preference for someone else to manage it.

Why most large enterprise will need virtualized data centers?

As more enterprise users move toward cloud-based applications, performance and speed of innovation (aka speed of change) become important. This has driven the desire to have some portion of infrastructure be located in a high performance, controlled IT facility, while other portions can be operated over a public infrastructure, creating a hybrid cloud network. This desire has created a new Cloud concept known as the hybrid cloud. Hybrid cloud is defined as having some portion of cloud infrastructure operating in a public cloud facility, and another portion operating in a private facility, usually to meet specific end user demands, such as security, location, accessibility and reliability.

Many large enterprises, including financial and retail enterprises, can build their own private cloud with private infrastructure they build themselves, or outsource it to a data center operator/integrator, who could own and maintain the space and connect it back to the main data center, where a rich set of other cloud services may be available. They could also supplement any public cloud offering with such a private cloud offer, thus offering a full hybrid cloud solution.
 

What is a Virtual Data Center?
The concept of the virtual data center is about both revenue and opportunity. A virtual data center is simply an extension of a larger centralized data center that offers similar services, with the added benefit of catering to more specific enterprise needs. By addressing large enterprises, perhaps through a customized virtual data center solution, data center operators can attract larger, more profitable end users while establishing a footprint in a new market segment. In fact, many Fortune 500 companies have already engaged in such strategies by building their own private networks using leased space and contracting companies to manage the network. This concept of the virtual data center has been shown to be popular in large metropolitan locations, where customers are spread out across larger distances.

Overcoming the network challenge

The key challenge is how to extend the network from the central data center to a virtual location efficiently. In fact, there are several ways to overcome this network challenge, including technology, operational simplicity and ultra-high performance. All of the key operational challenges concern space utilization, power consumption and the need for a high performance network. In this case, two key technological innovations enable such solutions to become reality: photonic integration and optical super-channels.

Photonic Integration

The value of photonic integration lies in replacing multiple discrete components with a single highly integrated optical integrated circuit, also known as the photonic integrated circuit (PIC). A key value of the PIC is that it reduces space and power while ultimately providing higher performance network capacity. It follows the same philosophy of integrated circuits in our laptops replacing discrete transistors that were much larger and consumed much more power. Ultimately the real value of photonic integration is that it takes the highest power consuming components (e.g., network side lasers) and integrates them into a small compact device that consumes significantly less power and space.

Optical super-channels

Optical super-channels are defined as a group of smaller, more granular optical channels that are bundled into a single, larger optical group that provides equivalent high performance, but also adds the simplicity of managing fewer circuits. For example, would you rather manage 50 x 10G fiber circuits or manage 5 x 100G fiber circuits?

If we agree that PIC based optical super-channels are the simplest and most cost effective way of deploying network capacity, the next question is one of reliability. Fortunately, PIC technology is so reliable that it features an expected Failure in Time (FIT) of more than 1 billion hours (Source: Infinera).This solution results in increased reliability with photonic integration and simplification with super-channels that deliver performance.

Networks that maximize the Return on Investment (ROI)

Leveraging both Photonic Integration and Optical super-channels will not only help drive greater network efficiency and operational efficiency, but also provide opportunities to increase ROI. However, 500Gb/s is a lot of capacity, but a particular customer’s forecasted need may be only about 100-200Gb/s. To solve this dilemma, the superchannel can be sliced whenever needed through software activation. Namely, since optical super-channels and photonic integration are tightly coupled, customers can simply activate the bandwidth they need, in 100Gb/s bandwidth increments, using a few simple keystrokes, similar to activating software on your computer.
 

Conclusion
As the data center market continues to evolve, large centralized data centers and smaller virtual data centers that are closer to the end users will co-exist. The concept of the hybrid cloud plays a role in that it addresses the need for large enterprises to keep certain mission critical resources close to them while locating other assets in large, more economical and centralized facilities. This solution also creates opportunity for data center operators to offer value-added services, from the basic virtual storage and computing services to the fully outsourced IT solutions that make the operators more indispensable to the enterprise. The critical element in this solution remains the network, which must be simple and efficient. From the technology angle, technologies such as photonic integration and optical super-channels may be critical to ensuring the deployment of simple, efficient and high performance virtual data center solutions.