Google Cloud Platform (GCP) is an excellent Infrastructure-as-a-Service (IaaS) solution that integrates with various Linux distros as well as with Windows Server through 2016. GCP (whose cost can be determined by contacting the company) has grown to 15 regions, with at least three zones each since we last reviewed it. Each zone has multiple data centers and is increasing as Google continues to build storage infrastructure. Google has expanded its network to meet the growing demands of its data centers and has had the recent addition of three undersea cables. Notably, one cable is to the western Pacific landing in Guam. This interconnects with 10 undersea cables connecting Asia, the South Pacific, Australia, Japan, and the US mainland.
Google’s global networking is a significant asset when considered with its wide choices in networking available to GCP users. Network setup is well-designed, easy to use, and provides a wide selection of assets including network infrastructure and virtual networking. All of this is now available in the newly designed GCP Console, which is intuitive and easy to use in its own right. Because of all these positives and more, we gave GCP an excellent rating in our IaaS review roundup, though GCP still didn’t beat competitor Amazon Web Services ($6,415.00 at Amazon) (AWS), which garnered our Editors’ Choice and might be slightly less expensive.
Setup and Configuration
Where once you needed to be a keyboard jockey to use Google Cloud, basic setup and operation is no longer the domain of the command line. Virtually everything can be accomplished through the new graphical user interface (GUI), and doing things such as creating and configuring a compute server is now reduced mostly to mouse clicks in your browser. This makes basic setup on Google just as easy as services with a simpler reputation like DigitalOcean ($1.00 at Digital Ocean) or Rackspace ($10,300.00 at Rackspace) .
To make things easier, Google provides a wide selection of preconfigured servers. Network configuration settings are also easily available, and you can set up servers through the user interface (UI).
That UI replaces nearly all of the functions that were previous only available through the command line. You access the menus and screens through your browser, and once your server is configured, you can access its GUI as well. When we set up our Windows 2016 server, for example, we used the remote desktop to manage the server just as if it were a local server, not one located hundreds of miles away.
It’s worth noting that GCP’s network performance is such that we saw no evidence of latency, even when using Windows Server through the remote desktop. While the remote desktop didn’t support our 4K monitor, it provided a screen that’s large enough to be managed easily.
GCP Performance
Since we last reviewed GCP, it has expanded its footprint significantly. As mentioned earlier, there are now 15 regions, each with at least three zones, and each of those has at least one data center, but most contain more than one. This footprint is being expanded with five more regions planned for 2018 and 2019. Google does not say how many data centers it has for GCP overall, primarily because the number keeps changing. It is fair to say that there’s probably a data center nearby wherever you are. For example, our testing was done by using the data center in Northern Virginia, which is only a short drive from where our review was being conducted. During testing, there was no discernible latency. There are GCP regions on every continent except Africa.
Each of these data centers is constantly being upgraded, with GCP currently offering servers with Intel Skylake processors. Google has said in previous announcements that it intends to upgrade to processors without the Spectre and Meltdown security flaws that have affected Intel processors. GCP’s internal network now runs at speeds up to 16 gigabits per second (Gbps). The network provides encrypted connections between regions using Google’s private global fiber network. All this adds up to a service with capabilities advanced enough to compete with enterprise-oriented services, including AWS and IBM Cloud ($9,100.00 at IBM) .
Testing GCP
As for the service itself, to test it I used Geekbench 4 from Primate Labs. Geekbench is a cross-platform benchmarking application that’s designed to treat all platforms the same, regardless of the operating system. For GCP, I used the minimum configuration needed for the benchmark: an n1-standard-1. This platform consists of a single Intel Xeon Virtual CPU (vCPU), 3.75 gigabytes (GB) of RAM, and a 10GB hard drive. The benchmark instance consisted of 64-bit Windows Server 2016 Enterprise. As was the case with the other services, we ran Geekbench from the remote Windows desktop.
This benchmark program runs several integer, floating point, and memory tests. With this benchmark, higher numbers are better. I didn’t try to formally test networking or storage performance.
GCP delivered an average single-core benchmark score of 2,827 and an average multi-core score of 2,659. The numbers are similar because were were only using one core. These results are very similar to those of the other cloud services also running Windows Server. While there were minor differences, it would appear that while the scores have improved somewhat over the previous benchmark for GCP, speed is not a significant differentiator for this cloud service.
Pricing and Contract
Google Compute Engine’s service-level agreement (SLA) guarantees at least 99.99 percent uptime. If your monthly uptime percentage is between 99.95 and 99.99 percent, then you’ll get a 10 percent credit. If it drops to between 95 percent and 99 percent, then you’ll get 25 percent. If it goes below 95 percent in a month, then you’ll get a 50 percent credit. Google provides pricing for each instance you create as you’re creating it. It bills on a per-second basis. For example, the benchmark instance would have cost $65.89 per month if I would let it run for one month (defined by Google as 720 hours).
I used GCP’s Pricing Calculator to price out a three-tier Ubuntu Linux web app. This was made up of two on-demand microinstances: the load balancer and the hosting website. The web server itself was hosted on a g1-small Linux instance. These were supported by a pair of on-demand, f1-micro Linux web servers for peak demand and a g1-small Linux disaster-recovery (DR) server.
For the database backup, I used 300GB and 4GB storage residing on Google Cloud Storage Standard. The 150GB database management system (DBMS) itself resided on a Cloud SQL D8 MySQL instance. The DR DBMS with 300 GB of storage made its home on a Cloud SQL D2 MySQL. The total monthly data transfer allowance, source to destination, was 440 GB, with a destination to source allowance of 1,140 GB.
As for tech support, I went with the minimum available support system. This helpdesk provided assistance via both email and tickets. In total, this simple eight-server web app package costs approximately $5,358.28 per year.
GCP is affordable and features robust performance even as Google has quickly added various data centers to support demand. The result of this massive growth is the unusual pricing for using GCP, which will vary slightly depending on location. GCP server farms in regions with abundant cheap energy will cost less to use as opposed to regions were the costs of electricity are higher.
However, Google’s aggressive rate of building data centers means that it’s much more likely you’ll find one near you. Using a data center that’s closer will reduce latency, both because of reduced propagation delays and reduced latency related to internet congestion and infrastructure latency. Depending on your app, latency may not matter to you. But, in general, you probably want to use a location near you.
It’s also worth noting—and Google makes sure to stress this—that there’s a lot more to GCP than just than just IaaS. You also have access to Knowledge-as-a-Service (KaaS), Platform-as-a-Service (PaaS), along with other services, including a wide selection of artificial intelligence (AI) and machine learning (ML) services, including Google’s well-known speech and vision ML features.
While this review was aimed at IaaS, GCP, like other cloud services, is much more than just IaaS. Chances are that GCP can do anything in the cloud that you’re likely to want done.