EC2 and GCE Pricing – It’s all About How You Slice It
Urs Hölzle stated in his recent keynote, Google’s “on-demand price is lower than the 3-year reserved price of most other providers”. Guess who he was referring to.
And his statement is true – in a way. Comparatively speaking, one of the most popular AWS instances is m1.large. For this instance, the comparison is very close: Assuming 24/7 runtime, here’s what we get:
However, this is only a partial picture. In the previous years, Amazon was traditionally lowering instance prices. Last year, Amazon took a different approach, and during the last few months, it introduced several new instance families.
One such family is the new “General Purpose” (m3). Its m3.large instance is a bit cheaper than m1.large, but it’s 60% more powerful in terms of CPU. In addition, it features two SSD disks should you launch this instance from an instance store.
The second family is the “Compute Optimized”, and its prominent member is c3.xlarge.
So now, the comparison table looks as follows:
For the comparison, we are assuming that Google GCEU is roughly 75% of Amazon ECU, and we take maximum Google price discount and Amazon 1-year heavy-utilization RI cost.
You can see that c3.xlarge is roughly twice as expensive as the n1-standard-2, but it’s a bit more than twice as powerful. M3.large is roughly 30% more expensive than n1-standard-2, but it’s also ~50% more powerful.
On top of all this, Amazon’s Spot pricing for EC2 can be as low as one-fifth of On-Demand pricing, making Spot instances incredibly attractive for applications and jobs that can be ended without any prior notice (AWS can stop these instances at any time).
So while Google’s pricing reduction is great for consumers, determining the best cloud option in terms of performance and cost is still no trivial matter.
To read about Amazon’s response to Google’s announcement, click here.
Sustained Usage Discounts – Too Good To Be True?
Doubtless, Google pricing structure is much more convenient for the DevOps folks. No calculations, nothing to think of in advance, all discounts are automatically applied. Almost a nirvana.
However, the new pricing is very far from simple. The discounts depend whether the instances were running concurrently or not, taking into account what Google refers to as “inferred instances” or the combination of multiple, non-continuous instances (of the same instance type) over the month, into a smaller number of continuously running or “sustained usage” instances. The discount also makes it extremely difficult to predict usage cost or to assess the transition cost from the internal datacenter, or another cloud to Google.
The credits resulting from the sustained usage discounts are applied at the end of the months, so no cost trending during the month is possible. In short – Google prices are not less complex than any others.
Clouds Don’t Like It Spikey
It’s interesting to note, that both Google and Amazon are encouraging sustained, concurrent use of instances. At both Amazon and Google, if you run, say 700 instances for 1 hour (which equals close to 1 month usage for 1 instance), you won’t get a sustained usage discount for the aggregated usage.
This is because both companies are always planning their capacity, and spikes are not good for them: they need to keep extra, idle capacity to accommodate spikes, so neither vendor is going to offer you any volume discount on spikey usage.
Google Cloud Storage – Awesome For Those That Use It
While storage is only 7% of cloud spend, 90% of storage is known to be used by just 10% of all cloud consumers. But for these consumers, Google storage is now the cheapest in town! Their standard storage is $0.026/GB – translating to anywhere from one-third to one-half of standard S3 pricing ($0.076/GB). Very significant indeed.