Opinion

AWS vs. Google Compute Engine — Who Trumps Whom?

  • author image

Google Compute Engine has finally arrived and is available broadly. Which should you go with? Here are four important considerations.

1. Pricing

Without a pay-ahead model, Google is significantly more expensive than AWS for comparable server workloads. The trouble here is that Google is confusing the story by saying they’ll only bill by the minute, with a 10-minute minimum. While this may make sense for edge cases, it’s usually little more than a rounding error as we are not aware of workloads that utilize sub-hour server time. In effect then, if you use a server for 500.5 hours, Google will bill you for 500.5 hours and AWS will bill you for 501 hours — this is merely a few cents difference and in our example below does not change the fact that Google is far more expensive. For any steady-state workloads based on an average of 720 hours per month (30-day month) customers will pay more to run the workload on Google’s Compute Engine vs. AWS.

Let’s take a real world example: If you use a typical small server (1-2 CPU units, 3.75-4GBs RAM), on Google Compute Engine it would cost $95.04 for the month, and on AWS it would cost $86.40. On the surface it’s not a huge deal — Google is only 10 percent more expensive. However, factor in AWS RI Pricing and that same server could be had for $48.33 on a one-year commitment (includes RI fee amortized) or $30.84 for a three-year commitment on AWS. That makes Google a whopping 97 percent more expensive on a one-year commitment and 208 percent more expensive on a three-year commitment!

2. Operating system flexibility

Google only allows Ubuntu or CentOs on its platform. While these are common flavors of Linux, they are far from the top-used platforms and are missing key flavors like Red Hat Enterprise Linux and the SUSE platform, among many others. Interestingly enough, the bigger issue here is the lack of Windows support. Windows has been, and for the foreseeable future will be, the number-one computing platform for enterprises. With server shares in the 70 percent+ range in recent years (source: IDC), Microsoft owns the enterprise server market. Not supporting Windows is not only a huge gap but it means the majority of existing businesses in the market for cloud services will not be able to buy from you. Quite an oversight.

3. Scale

We reviewed the “zones” in which Google Compute could be deployed. There’s no apparent geographic separation except the United States and Europe, and that’s not okay for folks that want to be in different flood or earthquake zones within the same country. Where are my West and East Coast options? Also note that an entire zone for Google is going to be down for maintenance for the next week! The only other option for geo-diverse deployment is out of country — Europe — and I don’t even have a secondary datacenter online. Not exactly confidence inspiring for those running enterprise IT operations, to be sure.

4. Functionality

There are major gaps in functionality on the Google Compute platform in addition to those called out above. Here are a few more:

  • There is no way to import or export a VM image into the platform.
  • Network configuration is very basic.
  • Complicated enterprise networks will be impossible to replicate in its current iteration.
  • There’s no load balancing on the compute engine! I can create multiple instances, but it gives me no way to use load balancing without running my own load balancer.

AWS trumps Google at this point. Top reasons: Google is more expensive for very basic features, is not appropriate for enterprise workloads and has no way to move existing content in or out seamlessly.

[Disclosure: 2nd Watch is an AWS partner and a Microsoft cloud champion.]

Kris Bliesner is CEO and co-founder of 2nd Watch, Inc. In his position, Kris is responsible for the company’s growth and sustainability. Kris oversees strategic development of 2nd Watch’s cloud-based software solutions, the technology development and customer technology solutions, the technology group and the technology trends for the company. Follow him on Twitter: @ThaBliez

 

Comments

By James

If you are running a 100 node Hadoop Cluster on EC2 that takes 100 min to complete, you pay for 200 hours since “each partial instance-hour consumed will be billed as a full hour.” However, with Compute Engine’s by-the-minute pricing, you can spin up 10x servers (1000) so it completes 10x faster (10 mins), and you’re charged less for less time.

By Kris Bliesner

James – you have keyed onto a specific point – pricing is definitely scenario based. On the surface using your 100 node example above with a small server (the example in my post) Google would cost $22 and AWS would cost $30 (on demand rates). However Google lacks not only a pay-ahead model but also lacks a spot rate. Using spot instances for small usage time workloads like this is very common. Under the current spot rate the AWS price would be $2.60 for the same $22 Google workload.

-Kris

  • Pingback: The Multi-Cloud: Who Really Cares? - Cloudyn

  • By Bobo Moreno

    70% marketshare for MS servers is just MS PR. That’s 70% from the PAID OS market, it does not include free software.
    AWS have multiple times stated that the vast majority of servers on AWS run Linux.

    Post Your Comment




    Leave another comment.

    In order to post a comment, you must complete the fields indicated above.

    Post Your Comment Close

    Thank you for your comment.

    Thank you for submitting your comment, your opinion is an important part of SandHill.com

    Your comment has been submitted for review and will be posted to this article as soon as it is approved.

    Back to Article

    Topics Related to this Article