Google says the way it runs its network is core to differentiating its cloud services.
Some people may think all major public cloud providers are alike in their network capabilities. Amazon Web Services, Microsoft Azure, and Google Cloud offer direct connections between data centers and their cloud through AWS Direct Connect, Azure Express Route, and Google Cloud Direct Peering (or VPN), respectively. At a superficial level, all systems seem similar. In practice, there are significant differences between cloud networks, and from Google’s view, that stems from how the network is run.
At Google Horizon, an enterprise customer event held recently in San Francisco, the internet giant highlighted the importance of its networking infrastructure to how Google Cloud operates. Cornelius Willis, lead for solutions and technical marketing at Google Cloud, told me in a discussion during the show that the “network is the cloud,” which reminds us that without the network, there is no cloud.
Google asserts that differences come from operating its own cloud network, which is the biggest backbone in the world, estimated to carry close to one-third of internet traffic. To support that point, it cited its heavy spending on capital, with about $9.9 billion in 2015.
For example, Google and its partners China Mobile International, China Telecom Global, Global Transit, KDDI, and Singtel funded a 5,592-mile-long fiber submarine cable named FASTER between the United States and Japan. The cable, installed by NEC, has a capacity of 60 terabits per second. Since Google controls this connection, it provides for better, predictable latency and bandwidth between its data centers, which in turn provides for better performance and high availability for applications and services. In a keynote, Urs Hölzle senior VP for technical infrastructure at Google, said that Google cloud will open about one new region a month in 2017, connected by its network.
Google's competitors usually rely on the public internet. Thus, while they do buy network capacity from partner carriers, it means sharing the public internet, which leads to potentially more unpredictability in data delivery. That implies there is a cost imposed on the cloud provider by the network provider, which may affect what enterprise pays per gigabyte for cloud data transfers.
A different take on hybrid cloud
The network also plays a key role in Google’s view of a hybrid cloud, which differs from those of companies that offer on-premises instances of cloud infrastructure software, such as an on-premises OpenStack or Microsoft’s upcoming AzureStack. With OpenStack or AzureStack, the on-premises data center runs a locally installed enterprise instance of its cloud infrastructure in the off-premises cloud. Applications may run on-premises or in the cloud, depending on resource requirements or for security and compliance requirements.
With Google's approach, enterprises do not run a miniaturized version of Google’s cloud on their on-premises data center. Enterprises instead use application infrastructure that's compatible with Google cloud. To create a hybrid cloud, workloads may run independently on-premises, in the cloud, or on both locations in concert.
If the application is split between on-premises and in the cloud, or distributed across Google’s cloud, then this form of computing relies on a secure, reliable network, and supports a view where most workloads eventually move to the public cloud. As a result, Google is not delivering a locally installed enterprise instance of its cloud infrastructure. Google's focus is on a hybrid environment that supports enterprises running the workload, as opposed to creating a hybrid environment that replicates on-premises cloud compute or storage infrastructure.
As complex and advanced Google’s networks are, access to its cloud services assumes good connectivity from end-user devices such as a tablet or a phone to the internet. Therefore, Google Cloud requires reliable networking across campus networks, Wi-Fi, or mobile networks – for which the enterprise is responsible.
These are elements that enterprise networking professionals will still strive to deliver. I had some bad WiFi access experience recently, and it reminded me that the last few meters of the network matter. If Google’s networking expertise can extend to improving that through better visibility into how the network runs end to end, it will be a much brighter future. But for workloads that runs within Google's control, the network plays an important role in tying all the workload elements together.