While reports vary these days on the exact level of demand for cloud computing by businesses, it's clear that there is fairly broad general interest if the features and cost are right and the risks are reasonable. In the short term, the organizations looking for effective new business strategies in the downturn are finding that cost is a compelling argument for cloud computing, especially for transitioning non-critical IT operations or to inexpensively harness cloud computing for resource intensive but low-risk tasks like software testing.
The real question for many IT departments is whether the cost of transition to an external compute cloud will be low enough to benefit from any medium-term savings. That process might be costly enough in itself to delay any significant ROI by outsourcing to the cloud. However, as we'll see below, there's now some indication that leading cloud computing vendors like Amazon are realizing that the current cost incentive just might not be enough, particularly for enterprises ready to commit for longer terms. As Microsoft gets ready to launch Azure and Google App Engine gets serious about the enterprise, expect pricing models to get even more competitive.
Compute Cloud Pricing for Amazon, Azure, and Google
Earlier this week Amazon lowered the costs of their Elastic Compute Cloud (EC2) reserved instances for organizations prepared for sign up for 1 to 3 year terms of usage (and who presumably understand their ceiling on steady state compute capacity.) This brings the cost of a continuously available cloud computing instance down by 30% to as little as 4.3 cents an hour (or a 3.0 cents/hour simple rate):
* - The Effective Hourly Rate is computed based on full-time (24x7) usage.
Note that this is just for computing power and some scratch storage. Dedicated long-term storage requires a separate cloud computing service such as S3 and other features as well, so this is just a comparison of raw core compute power. Back of the envelope calculations says that a cloud data center of 100 servers could be operated around the clock on EC2 for well under $5 an hour, or just under $38,000 a year all in (servers, power, cooling, facilities, basic operations), an impressive number. Note that Amazon's EC2 regular pricing for on-demand instances remains at $0.10/hour for Unix/Linux and $0.125 for Windows instances. It's worth noting that Amazon does not offer reserved instances for Windows.
For those planning their cloud usage, you can use Amazon's handy cloud services calculator to run your own scenarios.
Note that this makes Microsoft a hair cheaper (by $0.005/hour) than Amazon for Windows instances for the time being. When Azure launches, Microsoft says it will offer subscriptions that "provide payment predictability and price discounts that reflect levels of usage commitment." At that point, Microsoft will likely be much cheaper than Amazon for Windows instances, the latter whose commitment to Windows in the cloud seems fairly uncertain at the moment. On the other hand, you can't run Unix/Linux on Microsoft's cloud at all and that's not likely to change. Interestingly, both Amazon and Microsoft offer 3 and a half nines of service level availability (99.95%), making comparing the service prices even easier.
The last of the big three clouds is Google App Engine (GAE) and it has the exact same pricing as Amazon's EC2 for on-demand instances or $0.10/CPU hour. GAE is fairly restrictive in terms of what it will run in its cloud, supporting only Java and Python applications at this time, but it's also free for production use to a ceiling of 6.5 CPU hours per day and one daily gigabyte of transfer.
Conclusion: 5 Lessons From Today's Cloud Computing Value Propositions
Taking a look at all this, I've come away with five conclusions about the top providers of cloud computing today given their current pricing and feature sets:
- Amazon is currently the lowest cost cloud computing option overall. At least for production applications that need more than 6.5 hours of CPU/day, otherwise GAE is technically cheaper because it's free until this usage level. Amazon's current pricing advantage is entirely due to its reserved instances model. It's also the provider with the most experience right now and this makes it the one to beat with low prices + maturity. However, expect subscriptions from Azure to give it a run for its money when Microsoft's cloud platform formally launches in a few months (probably November).
- Windows costs at least 20% more to run in the cloud. Both Microsoft and Amazon offer almost identical pricing for Windows instances while Google App Engine is not even a player in Windows compute clouds. There are undoubtedly cheaper offerings from smaller clouds but they are less likely to be suitable for enterprise use, though certainly there are exceptions.
- Subscriptions will be one of the lock-in models for cloud computing. Pre-pay for your cloud to get the most value and you'll get great prices. But you'll be committed to providers for years potentially without a way to leave without stranded investments.
- Better elasticity does not confer major price advantages. GAE is one of the most granular of the cloud computing services, only requiring for you to pay for what you actually use (for example, you have to commit to at least an hour of compute time at a time from Amazon) but does not provide a major cost advantage for large applications.
- You can't pay more for better uptime and existing SLAs are not sufficient for important business systems. It's unclear why, given open questions about cloud reliability, why no vendors will offer differentiated service where enterprises can pay more for a better SLA. The best you can get right now is also the worst, or 99.95% uptime. This is about 4 hours of expected but unscheduled downtime a year. For business critical applications, this is still too much. This will end up being an opportunity for other vendors entering the space though I expect the Big 3 listed here will improve their SLAs over time as they mature.
I'm hoping to cover additional cloud computing providers and their pricing/value propositions more here this fall to round out our understanding of the competitive advantages that their cost models can provide businesses. In the meantime, expect few changes from the big players until after Azure launches and limited ones from the smaller players, which often use services such as EC2 to provide their own offerings. As Andrew McAfee recently pointed out, the transition to new economic models is long but usually inevitable and cloud computing is no exception.
What are your experiences with cloud compute pricing? How did it affect your decision to switch (or not) from internal infrastructure to the cloud? Please leave your comments below.