, , , , , , , , , ,

April 19, 2009

Clearing the Air On Cloud Computing by McKinsey & Co. Report

The leading management consulting firm McKinsey published a report at the Uptime Institute Conference which concludes that cloud computing makes no economic sense for large corporations. A detailed PDF presentation report is published on the Uptime Institute website.

Steve Lohr from the New York Times also published a short piece on the McKinsey report.

Cloud Computing delivers on customer benefits promised earlier by Network Computing (NCs), Zero Administration Windows, terminal computing and mainframes. Cloud Computing solves many of the problems companies of all sizes are facing today. Cloud computing is real, available today at a reasonable cost and not vaporware.

Our company has been aggressively Cloud Computing (not just POP3 hosted e-mail on a virtual UNIX or Linux machine) for nearly three years (yes, that is possible). After seeing the McKinsey report, I felt compelled to get on cloudy air.

In my previous life, I have also worked on Total Cost of Ownership (TCO) studies. I worked with leading industry analysts and sponsored total cost of ownership reports. Some of the products in which I was involved include Microsoft Exchange and Microsoft Windows NT. So, I am familiar with the variables involved and the complexity in evaluating TCO.

Unfortunately, the McKinsey report is based on pricing from a single vendor, the Amazon EC2 service, which may not be a good indicator of the real market price. The entire report and assumptions are based on one single marketing variable, the price of the Amazon EC2 service which Amazon can easily change in one press release. The real price competition is emerging now and will be fierce since leading vendors are making huge investments in Cloud computing.

The success of Cloud Computing will depend on other marketing variables beyond price, including technical features (product), customer awareness (promotions) and support of the application developers (ISVs) and system integrators (distribution channel).

In our limited experience with Amazon Cloud service on the Windows Server platform, all we purchased and configured was a virtual machine (VM) running an old SQL Server 2005 database server on the Windows Server 2003 operating system. All Amazon offered was a Virtual Machine which we had to pre-specify and pre-configure. There was no elasticity of computing power. This is really VM hosting and not cloud computing. Since our experience in November 2008, Amazon has dropped their prices. I have not checked but the earlier prices were high compared to provisioning a real server from a hosting provider.

We agree with the McKinsey report that current data centers will benefit a lot from aggressive use of Virtualization from Microsoft and VMWare. In fact, we take it to the extreme and do not even have a dedicated server for a domain controller. We run a Windows Server 2008 Domain Controller as a virtual machine on a shared host computer. We can easily back up this virtual machine on a 4GB USB flash drive ($15) in 7 minutes. Virtualization use has reduced our physical servers by 75%. Virtualization and cloud computing are not mutually exclusive.

The McKinsey report contradicts data shared by Harvard Business School Professor and Director of Intel for 20 years, David Yoffie. At the ACETECH Conference held on March 6, 2009 in Whistler, Canada, Professor Yoffie showed that the Cloud Computing data center cost for Google is only 25% of a large corporation’s data center. Google achieves cost efficiency through clever design of OS and application software, using standard low cost hardware and locating data centers near electricity generation plants.

Professor's Yoffie's data is consistent with a lot of data internally available to large players on how you can use low cost computers ($300 per machine) to redesign OS and applications for redundancy. If one computer goes out, there is no visible impact to users and there is no data loss. There is no need to debug the computer hardware right away. Just repair and reformat the computer. Software will do the rest.

Reliable Adaptive Distributed Systems Laboratory at the University of California, Berkeley published a report that is vendor neutral and provides additional data.

The McKinsey report also minimizes and does not discuss the need for corporate data centers to train and hire operators that can repair hardware, OS software (Windows Server 200x), server application configuration (e.g. SQL Server 200x) and the actual end user applications. Given our experience in dealing with data center technicians, first and second tier support staff lack the broad set of skills needed to run an application server. Given the pace of change, it is difficult for corporate data center operators to learn and stay current with these skills. Despite progress in standardizing data center procedures, the support documentation is obsolete the day it is published.

The entire computer industry will have to redesign server applications to support distributed computing and applications will have to be rewritten using the new Cloud APIs. The entire industry will have to deal with the fact that there will be only a few large buyers of hardware (Cloud Computing suppliers) that will wield huge purchasing power (probably no different than hardware OEMs such as HP, Dell, Acer today). Still advances in software will allow startups by recent Stanford graduates to set up new clouds and offer innovative services and prices.

The McKinsey report does not highlight some of the real end user problems with Cloud Computing. Real end user problems include single sign on for multiple applications as well as easy and affordable backup. Moving forward, cloud computing vendor switching costs and data migration will also be a concern.

The isolated examples of cloud outages and of end users being provided access to unauthorized data is no worse than someone losing a laptop or a data center employee walking away with company data on a USB Flash drive.

Updated on April 29, 2009 by Rajeev Agarwal