Subscribe / Unsubscribe Enewsletters | Login | Register

Pencil Banner

Google's cloud strategy becomes clearer with TensorFlow

David Mytton | May 19, 2016
With the announcement of Google's new machine learning hardware, its cloud platform strategy is becoming clear

For years, one of Google's biggest differentiators has been its global infrastructure. It has closely guarded its technologies, networking, and the data centers. This has allowed it to announce revolutionary features like the massive storage Gmail launched with, real time collaboration in Google Docs, and super-fast Web search well ahead of competitors.

These innovations all come from running highly efficient infrastructure at massive scale, something few of its competitors could match. However, with the rise of cloud computing, such advantages have started to disappear.

Google's response was to start productising its internal technologies, and Google Cloud Platform was born. Although very late to the party, Google has been rapidly building its portfolio of cloud products. It's differentiating through operational simplicity, and although it offers commodity services like compute and storage, it sets itself apart with managed services like databases, global load balancing, and container management.

Google Cloud certainly has some interesting compute features like sustained discounts, pre-emptible instances, and live migration, but in essence, anyone can offer compute as a service. The real reason to use any of the big 3 cloud providers (AWS, Azure, and Google) is their supporting portfolios.

Indeed, it's the services that couldn't be provided by anyone else that really stand out, particularly when it comes to large scale databases. Google's BigQuery is a great example of this because of the speed at which it can execute queries across huge datasets. Google provided an example of this when it demonstrated executing a 30-second query across 3000+ cores as a small job, and it has been backed up by customers like Spotify.

These services are built from the experience Google has gained building large scale software systems over the years. But they're experts at hardware too, and this is at the heart of Google's open source strategy.

As a tool to manage container clusters, Kubernetes was perhaps the first important cloud framework to come out of Google. It's an open source implementation ofGoogle's internal management systems released to the world for free. The goal is to standardise the APIs regardless of which platform they run on so. Of course, you can run your own Kubernetes cluster but Google is betting that you don't really want to do that kind of undifferentiated heavy lifting, and instead will use a managed service to do it for you. A service that Google Cloud Platform happens to offer!

This is a clever strategy to leverage open source to create standards whilst supplementing it with a service that can do it all for you more efficiently and with less operational overhead. Kubernetes and Google Container Engine were the first example of this but with the announcement of Google's new Tensor Processing Units (TPU) at Google i/o, we have another example. One that is even harder for competitors to replicate.


1  2  Next Page 

Sign up for MIS Asia eNewsletters.