Google Custom Machine Types: A Price-Performance Paradise or a Management Hell?

November 24 2015 | by Yoav Mor

0 Flares Twitter 0 Facebook 0 LinkedIn 0 Email -- 0 Flares ×

Google, while currently lagging behind AWS and MS Azure in the cloud services industry, is constantly innovating in both features and pricing models in order to position itself as a natural leader in the IaaS space.

With its latest innovation announced last week, Google Custom Machine Types, are trying to break the mold of predefined specs for virtual instances. Instead of offering a relatively small number (tens) of virtual instance types, “Custom Machine Types” opens up the full spectrum of server specs to Google Cloud’s users, enabling them to provision a machine with as many CPUs and as much RAM as they need (subject to some minor limitations). As many as 32 vCPUs and 208 GB of RAM can be provisioned.

Pricing of these virtual machines is determined by adding up two price rates: per-vCPU and per-GB RAM hourly price rates. On-demand price rates are subject to sustained usage discounts offered by Google Cloud on their predefined machines, and significantly lower preemptible VM rates are also available.

Learn about google custom machine types

Ostensibly, this is a great step in achieving optimal price-to-performance in the cloud. Application developers and operators know exactly how much compute and memory power is needed for for their application, and limiting them to provisioning predefined machines can often lead to underutilized resources. The option to custom-define a machine gives users the option of matching exactly the right performance that is needed for their application and pay just for utilized resources, and not for idle vCPUs or untouched memory space.

One must not, however, look just at the hourly price rate of machines when analyzing price-to-performance. Monitoring and keeping inventory of your cloud deployment gets far more complicated when the number of machine types increases from 20-30 types to 100s of types, each a bit different from one another. Monitoring such a heterogeneous deployment can create management complexities and inefficiencies, which outweigh the cost benefits in provisioning such machines.

It’s clear then that enterprises should not give a free hand to their cloud users in provisioning any type of machine they like, as this could result in cloud management hell and swelling overhead costs. Their IT departments should instead examine the performance requirements of their enterprise’s applications and derive a service catalog, limiting its users to provisioning a finite number of machines. Should the IT department come to the conclusion that vendor predefined machines don’t provide a price-performant solution, they can then specify a limited number of custom-created machines, which their users will be able to provision.

Final note
Google Custom Machine Types hold great promise in pursuit of the perfectly-optimized cloud deployment. The provisioning of such machines, however, should not be left to the end users who can in turn provision an endless number of machine types, creating a management and monitoring nightmare. Instead, IT departments should offer their users a service catalog, made up of a finite number of predefined machines, which can be either predefined from the vendor, or custom-defined by the IT department.

Learn more about balancing price and performance, and about monitoring and governance of your cloud deployment on Cloudyn’s website.

Start Optimizing Your Cloud Deployment

Start your 20-day free trial today to easily identify greater cloud efficiency opportunities and view your entire deployment in one place. Gain full transparency, accountability, and control.
start free trial

Connect with us
Sign up for our newsletter
  • альтернативный текст
  • альтернативный текст
  • альтернативный текст
  • альтернативный текст


SSO Login

Forgot Password?

No account yet? Register.

0 Flares Twitter 0 Facebook 0 LinkedIn 0 Email -- 0 Flares ×