1st impression on Google Cloud Platform
Last night there was a wonderful Spark meetup session that was hosted in Google Sydney office. One of the Google customer engineers Malcolm Holden presented Cloud dataproc —a managed Apache Spark/Hadoop service running on GCP.
This is probably the 1st time I watched a live demo for GCP. It was truely amazing that GCP managed to spin up a 100-node (1 master, 99 workers) Spark cluster within merely 90 seconds! Another cool feature is the build-in SSH accessibility. You can find a handy button after each node on the cluster management UI, by clicking the button you will be connected to the instance through a terminal (though a bit slow), saving the hassle of SSH key setup.
At the end of the presentation, we were provided with a couple of online resources regarding GCP, which I got chance checking into today.
The first question from new users for an unfamiliar platform is usually “any free tier option in place to play with?” GCP is no exception, actually, it provides quite flexible free tier schemes. Besides ‘standard’ 12 months free try period along with $300 free credit, a subscriber can play with any GCP product/service as she/he wishes. At the same time there is another ‘Always Free’ scheme for most popular GCP products/services that allow subscribers to have free usage up to predefined limits, on a per day or per month basis. For instance, each single day a free tier subscriber account can have following Google App Engine resources consumed free of charge — 28 instance hours, 5GB Cloud Storage, 1000 searches, 10MB search indexings, 100 emails. Any usage over Always Free usage limits would be charged against your credit, which sounds ituitive. Your free tier account expires either the 12-month period passes, or the $300 free credit runs out, whichever comes first. Read more on the free tier FAQ page.
The overview page under GCP documentation section is so informational that I suggest each beginner to have a read. The region, zone, and data center concepts in GCP are pretty much the same (or largely the same) as AWS. It looks currently GCP has only 3 regions —American, Europe, and Asia, one for each continent.
There are different access granularities for resources. For instance, networks are global resources, while external IP addresses are regional. Disks can be accessed only within the same zone. After all, it’s probably not a good idea to attach disks to instances located in a different zone, performance counts.
Another interesting concept in GCP is called ‘project’ — a namespace that groups multiple GCP resources together. Each project has a unique ID, which would be retained even after the project is destroyed. It makes accounting clear as one project can only be associated with a single GCP account.
Of a handful of tools provided, GCP console is usually mentioned firstly as that’s the face of the whole GCP product family. My first impression of GCP console was very positive :) It’s well organised and precise, if you were sick of AWS’s ‘hundreds of icons’ approach, like me.
Of no surprise, GCP provides gcloud CLI to give users access to a list of commands to manipulate GCP resources. It will be quite useful if one would like to automate building pipeline process.
Since pricing and usage tracking are so important that GCP provides a price calculator to give subscribers a preview and have a brief idea from budget perspective.
The first impression I got from Google Could Platform was — precise, open, great documentation. Though not having comprehensive services and global coverage compared to its major competitors, with ambitious growing plan and considable advantages on data & machine learning capabilities, GCP is going to play a very critical part in the public cloud space, provide solutions to help businesses achieve less-overhead application deployments.