5 Reasons Why Google Data Centers Are Running Like Giant, Efficient Computers

Performance Equals Results

Google has spent tons of time solving problems associated with running global-scale services. In many cases, Google has had to invent technologies to address specific challenges simply because no one else has attempted to build services at the scale that it's focused on.

In the past decade, Google has been developing its own networking software and hardware, and that has led to a 100x increase in data center capacity, Amin Vahdat, Google Fellow and technical lead for networking, said in a blog post Thursday.

This problem-solving mindset has helped Google build a portfolio of proprietary technologies and techniques that continue to fuel its success as a company. Following are five examples.

1. Using Clos Networks

Google, in designing its data center networks, follows Clos network topology, an approach that emerged in the 1950s as a way to handle telephone call switching. Google uses commodity switches and ties them together using software, enabling them to collectively act as a larger logical switch.

"We use a centralized software control stack to manage thousands of switches within the data center, making them effectively act as one large fabric," said Vahdat in the blog post.

2. Building Its Own Hardware, Software

Some vendors use products and technologies from other vendors to deliver products and services, which is one reason why you always hear Oracle Chief Technology Officer Larry Ellison talking about how Salesforce.com uses his database to run its cloud.

Google has adopted a different approach, preferring instead to build using its own software and silicon from hardware vendors, Vahdat said in the blog post.

Google has also cooked up its own custom network protocols that are optimized for its data centers, as opposed to using standard Internet protocols, according to Vahdat.

"Taken together, our network control stack has more in common with Google’s distributed computing architectures than traditional router-centric Internet protocols," said Vahdat in the blog post.

3. Adhering Early On To Software-Defined Networking

Vahdat said Google has been "deploying and enjoying the benefits" of software-defined networking for the past 10 years, getting an early start on an industry trend that's now in full swing.

Andromeda, Google's SDN network virtualization stack, and B4, the company's data center wide area network, are examples of network architecture that's designed around SDN principles, said Vahdat.

4. Thinking Differently

Yes, "Think different" is Apple's marketing slogan, but it applies to Google as well. The search giant has adopted tactics that aren't industry norms, because that's what it deems necessary to solving challenges no one else is attempting.

Google is constantly re-engineering and redeploying its data center network in order to scale its services for more and more users, and that's no mean feat, Vahdat said in the blog post.

"Our approach to networking fundamentally changes the organization of the network’s data, control, and management planes. Such a fundamental shift does not come without some bumps, but our operations team has more than met the challenge," said Vahdat in the blog post.

5. Building For Speed, Performance

The heavy-duty engineering Google has done over the past decade has enabled its data center network to run with ever-increasing efficiency, said Vahdat in the blog post.

The current generation of Google's data center network, called Jupiter, offers in excess of 1 Petabit/sec of total bisection bandwidth, or the speed data is transmitted between two networks, said Vahdat.

According to Vahdat, this means 100,000 servers could exchange data as speeds of 10 GB/sec each, which is "enough to read the entire scanned contents of the Library of Congress in less than 1/10th of a second."