I’ve recently posted an updated version of the cloud IaaS vendor comparison. For me, there are two main uses for any kind of comparison data: the first one is to pit providers against each other, checking to see who is going to offer cheaper servers, better support, and so on. This is very useful for newcomers to the cloud, people who are looking to change providers, and for other similar uses. The second, less obvious application of any comparison data is market exploration, looking at how the different companies position themselves on the dimensions specified and trying to find market trends from this positioning. Just a quick recap: the comparison includes 16 companies, broken down into “Top-of-Mind” companies (the leaders, the ones we immediately think about when talking about the cloud) and the “Upstarts” (companies that are less well-known, but still provide good service). I created two subsets of dimensions, one of “Cloud Promises”, covering the main promises of cloud computing (cost optimization, scalability and automation, and flexibility) and one of “User Concerns”, dealing with the greatest concerns users have when moving to the cloud (security, vendor lock-in and reliability).

There are, in fact, several different interesting studies that we can extract from the data, but first some care must be taken. Our first step for analysis is going to be data normalization (not to be confused with database normalization), so that we can compare the companies without a single dimension completely distorting our analysis. Normalization was done using a simple standard score; discrete dimensions were placed on a 0-2 scale, where 0 is the lowest (worst) grade and 2 is the highest (best) grade. From here, these two simple charts compare the average scores of “Top of Mind” companies with the “Upstarts” (if you’re wondering about the scale, it’s in z-scores):

From this first chart, we can see that the “Upstarts”, that is, the smaller companies that are fighting against the leaders in this market, are on average charging lower prices and offering a higher degree of customization of virtual instances (more “Instance Types”); the larger providers, as expected, have more robust offerings, bringing to the table more data centers, better monitoring tools and APIs, and a greater variety of pricing plans.

On the “User Concerns” we can also see some interesting polarization: the smaller providers seem to be focusing more on offering premium service (through extended service hours and more contact channels), more aggressive SLAs and the possibility of uploading your own VMs to the services. The larger ones have much better security ratings, however:

These charts only look at the averaged data, so let’s look at the details.

  • On the “Cost Reductions / Optimizations” section, Lunacloud comes out in front. They have the cheapest servers I’ve found, at roughly US$ 46.00 per month, as well as the cheapest storage cost and the second cheapest outbound data cost (no charge for inbound).
  • On the “Scalability and Automation” section, Amazon and Softlayer come out on top, offering rich APIs, full scale out and scale up functionality, and good monitoring tools to round it out.
  • On the “Choice and Flexibility” section, AT&T wins by sheer numbers: 26 global datacenters to choose from, with fully configurable instances. And even if they only support two operating systems out of the box, you can upload your own VM images with whatever software you want.
  • On the “Security Features”, many companies are mostly tied, with datacenters with all the required certification and some security features on their offerings – but these are far from complete.
  • The same thing happens with “Ease of Migration”. Most companies, especially those employing VMware technology, are allowing for easy VM upload and download, simplifying the life of IT departments.
  • Finally, on the “Reliability” section, we have Rackspace, OpSource, Softlayer and Hosting.com taking the top spot, with services that have been running for over 5 years, very aggressive SLAs, and extensive support.

The rise of VMware is an especially interesting trend. While some of the large providers are spending time and effort trying to create open source cloud standards, such as OpenStack and CloudStack, VMware is quietly taking over the market. In the group of providers checked, there are more companies using VMware technology than any other cloud standard. This trend could end with VMware becoming the true cloud standard, and it does, in fact, make sense: many midsize and large companies already use VMware internally, so the migration from internal data centers to the cloud becomes much easier if the cloud provider offers technology they are familiar with. All providers that work with VMware are also offering the possibility of clients uploading their own VMs, making any transition even simpler.

Finally, we can look at who stands out from the pack. For this, a simple analysis is to simply sum the normalized scores for each provider, assigning an equal weight for all dimensions. In this case, the top providers with respect to the “Cloud Promises” are Softlayer, Opsource, Rackspace, Amazon, and Lunacloud. These providers all have low prices, good APIs and monitoring tools and many instance types and datacenters to choose from. Looking at the “User Concerns”, the top providers are Softlayer, Opsource, Hosting.com and Tier3. Rackspace and Amazon are solid, but don’t stand out so much here. The top providers on this category all offer excellent customer service, aggressive SLAs, and have all the security certifications on their data centers. They are also services that have been online for 5 years or more, so they have a longer track record.

There are many other interesting trends that can be extracted from the data. I’m providing the full Excel spreadsheet, already with the numeric transformations and the normalization, so that anyone can work with the data if you want. Rank the providers, add (or remove) dimensions, add other providers, place weights on the dimensions to change the scores or simply change the data around. If you come to any other interesting conclusions, please share in the comments.