Balakrishna Narasimhan
The venerable management consulting firm, McKinsey & Company, released a thought-provoking analysis yesterday on cloud computing economics. The piece has generated a fair bit of attention because it's been taken to mean that migration to cloud platforms is actually more expensive than what large companies currently spend on their own datacenters.
As usual, the problem is not in the analysis or the research but in the question that is being asked. The question that the McKinsey analysis answers is about the comparative economics between running your datacenter on your own hardware vs. running it on Amazon's hardware (offered as a service). We aren't going to question their analysis or numbers (we'll leave that to experts like Vinnie Mirchandani), but we also don't think this really answers the question about what cloud platforms can do for a business.
Cloud platforms exist at three levels (Click here to enlarge image)
At the lowest level, infrastructure-as-a-service is purely computational power for rent, which is what services like EC2 offer. It abstracts the physical infrastructure but you still need to do the work of mounting a database and an app server on the infrastructure, building and maintaining your app, etc. Therefore, the only savings are those that come from the delta between how efficient your datacenters are vs. those that Amazon runs. As you talk about large, well-managed datacenters that are operating at scale, it's plausible that savings are not significant.
It's at the next level, Platform as a Service, and beyond, that we start to see significant savings. Once you move up the stack to PaaS, there are significant savings because you no longer need to run a datacenter (physical or virtual as in the Amazon case) or maintain infrastructure software (database and app servers). Within our 150+ customers, we see savings of over 30% on operating costs and 2-3x improvements in time-to-market when building on cloud platforms. For example, for a publishing client, we built a custom application that automated the entire publishing process in less than 6 months. Their estimate for doing this using on-premise platforms was over 3 years. In terms of ongoing cost/productivity improvements, they have estimated a 50-75% reduction in the time and effort it takes to add new products. Additionally, since the application is built on the Force.com platform, upgrades are seamless and the platform gets better over time, all for no additional cost.
At the highest level of the stack, the benefits get multiplied further, since you get all the benefits of PaaS, plus you get freed from 22% maintenance and costly (to implement) upgrades every 3-5 years. The savings have been well documented: 25-40% in terms of implementation costs (by freeing yourself from the clutches of the dreaded Globals SIs) and operating cost savings, e.g.,50%+ savings running your mail on Google vs. Exchange.
Cloud platforms provide savings at each layer of the stack, and McKinsey's analysis focuses on just the lowest levels of the stack, thus missing most of the savings potential.
We have seen the benefits of cloud platforms first-hand at over 150 customers, including companies like Avago, Genentech, Japan Post, Qualcomm, Starbucks and Dolby. Once customers experience the benefits of cloud platforms - quantifiable savings, rapid time to value and innovation that drives the business, they seldom want to go back. This is why 90%+ of customers plan to increase their spending on cloud platforms. In these economic times, there is no greater vote of confidence for cloud platforms than that!
Thank you and well done. This is a great positioning of the players and their platforms & services.
ReplyDeleteThis is a fine analysis. I'm going to link it to my own blog. I had the job of commenting on this article before I saw it (ugh -- that's never a winning proposition) and I was sure that my disagreement with it would be that we were looking at cloud computing from different perspectives. But I certainly didn't offer so elegant an explanation as offered here.
ReplyDelete"plus you get freed from 22% maintenance and costly"
ReplyDelete>>I am not sure I get this. In case of PaaS, don't you have to pay monthly fees?
Your argument of faster time to market is right. However, I am not sure you have proven that TOC is lower for PaaS v/s on-premise solution.
Peter and Amy, thanks for your comments.
ReplyDeleteJack, excellent point that we haven't proven that a SaaS or PaaS app is cheaper, in this piece.
However, our experience with our customers is that there are substantial TCO savings from moving off on-premise apps to SaaS. You still pay the subscription fee but it's typically a lot less than maintenance + HW + ...
One additional savings comes from the fact that you have one plateform to build application and not a bunch of different technologies to build them. They are more consistent and it is easy to reuse code or add one new tab to an existing application instead of devlopping a new one
ReplyDeleteI agree that the real ROI and saving can be measured from the medium level of PaaS and up. Are there any references or analysis that refer to the cost and ROI when moving legacy systems (as Mainframe) to the cloud?
ReplyDeletealso there is a cost saving from not have to employ an army of people to a. develop the app b. maintain the app + infrastructure (unix admins, dbas etc)
ReplyDeleteThe other thing not in the McKinsey scenario, as you note, is "infrastructure management and provisioning," which you do not examine here - but others certainly will. Savings can be significant.
ReplyDeleteGraeme
@GraemeThickins
The problem with this response is really an indication of the softness of the field. This article talks about services and their value as a response to the oversight of a study about infrastructure clouds. This is as coherent as talking about MySql vs. Oracle, concluding MySql is cheaper and having someone say the study wasn't complete as someone hadn't considered the value of Apache and Websphere in the study.
ReplyDeleteMy kingdom for a field consistent vocabulary of platform, clouds, services, etc..
@mengerin agree with you that there is a problem of definition consistency in cloud computing.
ReplyDeleteHowever, the reason for our response is because the true benefits of cloud computing come in when companies replace their datacenters by building their apps on PaaS or using SaaS apps.
In fact, a great answer to your MySQL vs Oracle scenario, is to move to PaaS or SaaS and stop worrying about database administration/maintenance!
Interesting angle.
ReplyDeletePersonally, I think that the Iaas savings are much bigger than you assess, largely because enterprise operations is too far away from the business and so has not been able to get the economies of scale for any type of infrastructure. This issue is compounded as ops staff are motivated to make the dc as complex as possible.
Additionally, I think that it is quite feasible to put provisioning into the app dev. process, for production, test, dev and training systems. And the pricing advantage of Iaas, gives the AD business the win to ensure that they make it work.
Saas works, as is, but doesn't integrate well into the rest of the value chain. Paas may work, but I think that the frameworks are too immature for general use at the moment.
I've not managed to get hold of a copy of the McKinsey report, but from what I've read they are very optimistic about expected asset utilisation: IT as an org is rubbish at capacity planning and demand management, so a really good average utilisation would be somewhere near 15% for the server estate.
I think that 'Private clouds' are just another bid by IT to expand its empire. Which enterprise needs to fund a proprietary alternative to AWS?
Although there are current compliance and legal accountability issues with data location, I don't think that there are many technical real threats: it would be nearly impossible to find the right VMs to highjack while processing data and then to piece it all together. All data out of RAM should be encrypted (there's a learning curve here, but it's not rocket science).
If Berkeley/Hamilton's analysis of the cost advantage of megadatacentres is close, it blows out the tax benefits of owning your own kit and the pay as you go model can create sufficient cost transparency to destroy eliminate hidden costs in ICT delivery.
Question: wouldn't savings diminish for SaaS for companies that are large with complex business processes?
ReplyDeleteActually, its exactly the opposite! Large companies have the most to gain from using Software, Platform, and Infrastructure as a service. Here's why:
ReplyDeleteFirst is TCO: There's no such thing as an "at scale" data center. Even the largest companies in the world are sub-scale relative to the consumer internet, which is the basis of the cloud platforms provided by Google, Salesforce, and Amazon.
Second is business benefit: Custom applications are cheaper and faster on the cloud. You're right that larger companies have customized business processes-- but building those custom apps on an on-demand platform allows a large company to focus their resources on what's specific to their business... not the underlying infrastructure!
For examples of large companies who are using the cloud to power their business, check out these case studies.
"90%+ of customers plan to increase their spending on cloud platforms" ... Amen to that!
ReplyDeleteObviously companies like Microsoft can no longer ignore the swing towards the cloud - if you can't beat em join em right.