Google Compute Engine – What’s in Store?

Here is a closer look at the service –

Google Compute Engine offers full virtual machines running either Ubuntu or CentOs. Customers can launch VMs based on 1,2,4 or 8 virtual core instances with 3.75 GB of RAM allocated per core. Each VM comes with temporary or ephemeral storage that may not be suitable to persist durable data. To add durability, GCE offers persistent disks that can be attached to a VM. Periodic snapshots taken from the persistent disks can be stored on Google Cloud Storage which an object storage.

GCE’s network isolates each VM to avoid "cross talk" among the shared tenants. Customers can also configure a virtual firewall to restrict the traffic to the public facing VMs. VMs can be associated with a static IP address which can pointed to a custom domain.

The tooling experience is primitive which is mostly based on command line tools. Google has partnered with RightScale, Puppet Labs and OpsCode to simplify DevOps. But currently the provisioning and life cycle management of the VMs is limited to command line.

With the investments made in Google App Engine, Google wants the customers to run consumer facing web sites in GAE and target compute intensive workloads on GCE. Though there are no restrictions in hosting LAMP or Java websites on GCE, Google is encouraging the customers to run batch processing, analytics and HPC workloads. Urs Hölzle, Google’s senior vice president of technical infrastructure demonstrated a genomic app that exploited 600,000 cores powered by GCE. Google has invested in massive compute power waiting to be exploited.

The key differentiating factors of GCE happen to be scale, performance and cost. Google is relying on their secret sauce of scaling based on inexpensive hardware. They are comfortable dealing with high volumes and low margins and it happens to be a proven business model for Google. Taking a shot at Amazon, Urs claimed that the underlying infrastructure of GCE is designed to be highly performant while offering significant cost benefit to the customers. He claims that GCE is at least 50% cheaper than the competition. Based on the published GCE pricing details, this seems to be correct. GCE does offer more compute per dollar when compared to Amazon EC2. Below is the pricing chat of GCE.

It is interesting to see that Google has announced the language bindings for the GCE API in popular languages including .NET, Java, JavaScript, Go, PHP, Python and Ruby. The API seems to follow the Google conventions and reflects the semantics of the remaining product APIs. Since Google Cloud Storage API is compatible with Amazon S3, I expected GCE to come close to the Amazon EC2 API. That would have given some advantage to Google and its developers by enabling the reuse of the tools and the integration with the thriving AWS ecosystem.

GCE will be available in Americas during the preview but will be launched in other regions including Europe and APAC in the future. During the preview, Google may approve access to early adopters whose workloads demand 100+ VMs.

Though it is early to draw conclusions, Google’s entry into IaaS space is an important milestone in the rapidly growing Cloud market. Earlier this month Microsoft announced the IaaS offering on top of Azure. With PaaS players switching to IaaS, it does raise interesting questions about the monetization of PaaS.

- Janakiram MSV, Chief Editor, CloudStory.in