The unbearable lightness of being: Trying to know in advance how much a cloud service costs
 |
| “Sometimes you make up your mind about something without knowing why, and your decision persists by the power of inertia. Every year it gets harder to change.” ― Milan Kundera, The Unbearable Lightness of Being |
From Huan Liu’s Blog we found some revealing cost comparisons of Terremark, RackSpace and GoGrid versus Amazon EC2. There are various sources for this type analyses that convey a deceiving first impression that making cloud cost comparisons is easy, It is not, Most providers made it a rocket science.
The Rackspace cost comparison with Amazon EC2 entry is from January 25, 2011. Interesting RackSpace and GoGrid charge only for RAM and not for CPU, as the CPU is shared proportional to the memory it uses. Did you know this? I didn’t. So if you use little RAM 1.5 GB, the ratio of the RackSpace Cost / EC2 cost is 17% at low RAM and it jumps 191% if the one selects 16Mb.
RackSpace was and still is a hosting provider like the wolf from the Little Red Riding Hood They attract subscribers to their cloud, then when a serious usage occur, the prices jump double the costs of the 100% cloud providers. This is natural, as RackSpace, same as Go Grid, cannot offer competitive cloud prices, unless they erode their own co-location and hosting customer base.
Another entry in this blog, Comparing Cloud Providers on VM Cost shows RackSpace and Terremark cost ratios with an equivalent EC2 cost from 17% to 50% ONLY when the memory usage is low (0.25 GB to 1.5 GB maximum).
Terremark cost comparison with Amazon EC2 has in some cases (high RAM versus minimum CPU allocated) cost up to 3x to 5x the EC2 costs. GoGrid has similar prices as EC2 at low RAM, but it jumps 4x in high ram configurations.
The above data prove clearly three bullets
- Former hosting providers converted to cloud providers have an intrinsic obstacle of bringing competitive cloud offerings, as they could erode their lucrative hosting business
- The $25 million in cloud revenues minimum requirement for admission to Gartner's Magic Quadrant for Cloud Infrastructure as a Service is not fair. It is virtually impossible to discern how much income GoGrid, RackSpace and Terremark have from cloud versus hosting services
- It is sad how little transparency most cloud and hosting companies offer to predict a budget.
Amazon EC2 so called simple monthly calculator to predict billing is nothing by simple. You have to know Number of Elastic IPs, Elastic IP Non-attached Time, Number of Elastic LB, Data Transfers In and Out, Regional Data Transfers, and so on and on and on. And this is just EC2. Then you have to know CloudWatch, in case you use it, asking again details I have no idea, because as an advance estimator, I supposedly do did not use AWS before. As most people, I blush to admit that I do not know how to write an AMI (along with 95% of the people using AWS who are not developers).
Comments
I don't think that most of the hosting providers trying to go cloud are particularly concerned about eroding their core offerings, especially since the larger managed hosters have generally made money on services rather than on the equipment. And the cloud represents a market expansion into a whole host of new use cases. But I do think that they price at what they think customers will pay, usually based on some price-comparison to Amazon. The VMware ones also have the "VMware tax" to deal with (i.e., the relatively high licensing cost of VMware).
Importantly, whether you are talking about Amazon or anyone else, the prices listed publicly are often not what customers actually pay. (For readers who might not be aware of this: Amazon will offer very different discounting structures on an Enterprise Agreement contract.) So comparing costs using public figures often doesn't accurately reflect the real-world relative cost of different competitors.
In the previous Magic Quadrant, providers had two options for qualifications -- meet a fairly high revenue number for all hosting-related services, or meet a much lower number for pure cloud. Doing it that way allowed us to include a bunch of smaller interesting providers who had a substantial cloud business but weren't huge hosters.
The qualification survey for the MQ required vendors to tell us (under an NDA) their cloud revenues, run rate, customer count, and a number of metrics around VMs, along with a variety of other hosting metrics.
Building an AMI, by the way, is a trivial exercise. (Or at least, I found it to be so, although admittedly, I spent the first several years of my career as a Unix-flavored systems programmer / systems administrator.)
I am a fan of the MQ research paper - probably the best analyst source so far for IaaS evaluations. But it is made by humans, not gods, and therefore it can be improved.
The $25M revenues threshold is biased toward hosting companies doing some cloud implementations in additions to the hosting business. We just discovered that the revenues from cloud services are blurry and hard to predict. I"ll surmise just signing an NDA, one can not know the truth of what is cloud and what is hosting, because the challenge of auditing such information.
Perhaps one can do two MQs: one above $25M and one below $25. This way one can capture all emerging new promising IaaS providers, with the highest growth
hosting forums
web designing company