I’m pretty sure that you would agree with me if I told you that your enterprise network will just be getting bigger and bigger over time. As the person with the CIO job, it’s going to be your job to find ways for the company to keep building out its network without going broke. Got any ideas on how to go about doing this?
How Google Created A Very Large Network
Google has been in the business of creating very large networks for over a decade. It was a decade ago that Google took a look at how they were building their network and decided that this was not going to scale well. They started out like everyone else going to the big hardware vendors and purchasing high-end equipment from them. They quickly learned that they were facing cost, operational complexity, and limited scale issues that were endemic to datacenter networks a decade ago.
The problem that Google was running into was that the networking equipment that was available to them at the time did not exactly meet their needs. Ten years ago, what they found was that the cost and operational complexity associated with traditional datacenter network architectures was prohibitive. Critical issues to Google such as maximum network scale was limited by the cost and capacity of the highest end Cisco and Juniper switches available at any point in time. Network control and management protocols targeted autonomous individual switches rather than preconfigured and largely static datacenter fabrics. Most of these features were not useful for datacenters, increased cost, complexity, delayed time to market, and made network management more difficult.
The choice that Google made then was to go out and build the network hardware that they needed. Now this was not the first time that they had tried this. The effort was inspired by the company’s success in using commodity servers for high-performance computing. One thing that we CIOs need to keep in mind is that the old way of doing things may be coming to close. Some corporations have begun to experiment with commodity hardware and other alternatives to replace conventional switches in their data centers.
Google’s Lessons For CIOs For Creating Large Networks
Let’s face it – Google currently runs some of the largest data centers in the world. With its current network, each version of their network has a different name and current one is called Jupiter, Google has succeeded in operating a network running on off-the-shelf switching components that has scaled to more than 1 petabit per second of total bisection bandwidth. This means that means each of the 100,000 servers in a warehouse-size data center can communicate with each other in an arbitrary pattern at 10 gigabits per second. No, we probably won’t build networks to run this fast, but Google sure looks like they could teach us a thing or two about how to do it.
One of the things that CIOs need to realize is that Google didn’t just create the network that it has today. Rather, over the past decade, Google has created five generations of networks. Some core tenets remained the same, though, in each of the five generations of networking equipment. They’re all characterized by merchant silicon which is off-the-shelf chip components designed for networking hardware which can be bought from other companies. Finally, all use sophisticated centralized control software for managing the complex network.
As CIOs start to explore using off-the-shelf switching components to create their datacenter networks, they are emulating the way other large-scale Web companies such as Facebook Inc. are creating their networks. The way that Google went about building their custom network is different from many companies because they used merchant silicon whereas others have bought so-called white box switches from Asian suppliers which come with the merchant silicon already embedded. The key point is that even Google recognizes that others are starting to encounter the networking challenges faced by large-scale Web companies. As CIOs begin to handle Big Data and analytics in their own data centers, they’re faced with providing a data center network that can scale to handle the traffic. Understanding how Google has tacked these same problems is a great place to start as we create our own datacenter network solutions.
What All Of This Means For You
As the person with the CIO position we understand the importance of information technology and we understand that our enterprise network will just keep growing. What this means for us is that we are going to have to be making some tough decisions about the best way to grow our network. Google has faced the same challenges that we are facing and they can provide us with some guidance.
When Google first started growing their network a decade ago, they did it like everyone else – they bought hardware from the top vendors. However, they quickly learned that the boxes that they were buying contained features that they were not using and could not scale to keep up with the Google network’s growth. Google then started building their own datacenter networking equipment using off the shelf parts. Other firms have followed but what they are doing is buying generic boxes from overseas vendors.
As a CIO you need to take a careful look at what Google is doing. They make a good point that a datacenter environment is a unique place where a lot of the functionality that comes in today’s commercial networking equipment is not needed and in fact makes things more complicated. You may not go all the way and decide to build your own equipment like Google does; however, purchasing simpler gear from overseas vendors might be a great way to save both money and network complexity.
Question For You: Do you think that the cost savings makes it worth using off name products in your network?
P.S.: Free subscriptions to The Accidental Successful CIO Newsletter are now available. Learn what you need to know to do the job. Subscribe now: Click Here!
What We’ll Be Talking About Next Time
Wow – that Facebook, Twitter, Instagram, etc. sure got really big overnight, didn’t it? One idea that has been running through almost every CEO’s head as of late has been the concept of creating something like Facebook, but restricting it to only exist within the company. The idea that employees could quickly and seamlessly exchange ideas is a great concept, but will it work in real life?