Author Archive

How Network Effects Are Likely To Power The Cloud

November 6, 2008

The recent posts about the economics of cloud computing between Nick Carr and Tim O’Reilly (here and here) and the panel at this week’s Web 2.0 have created a lot of buzz. The central question is of great consequence: will the emerging “cloud” operating system generate the monopoly rents and industry control that Microsoft enjoyed with Windows? For the sake of argument, let’s assume VMware is leading in private enterprise clouds and that for now Amazon leads public ones with Google, Yahoo, Rackspace, and Microsoft as contenders. It seems likely so far that Microsoft will have a significant advantage in private clouds, though not to the same extent as with Windows. Public clouds seem years further out, so they’re harder to handicap.

But at the center of the argument is whether dominance in either variety will come via Web 2.0 style “harnessing collective intelligence” or the more traditional “network effects.”  I believe we will see Microsoft emerge as a leader in private clouds in 3-5 years and it will be on account of the more traditional network effects.  Market share won’t accrue to the leader by virtue of capturing more information every time someone uses the cloud.  Instead, it will accumulate the way Windows steadily accumulated application and device support over time.

The two critical success factors in cloud computing: virtualizing the hardware and managing the software components

There are two key assets to leverage success in this market. VMware and Microsoft are likely to share self-reinforcing leadership to the first. That is the ability to make a sprawling and heterogeneous collection of servers, storage, and networking look like a single machine. Through automation interfaces, this capability dramatically changes the economics of administering data centers. The other key asset to leverage for success in this market is the ability to combine infrastructure and applications management. That is the critical requirement for turning an IT operation into a private cloud that can deliver rock solid online services.

Leveraging network effects to make the hardware infrastructure look like one big machine

As Tim explains, the cloud has multiple layers. But the bottom-facing utility layer sits on the hardware infrastructure and uses this generation’s equivalent of device drivers to make it look like a uniform pool of resources to the software above this layer. This is VMware’s strength by virtue of its lead in virtualization. Market share begets network effects at this layer in terms of device support, such as the widest range of storage devices and access to all their unique features. However, over time, it’s hard to imagine how Microsoft will fail to leverage its Windows Server and associated Hyper-V unit volume to achieve similar device coverage. So although the utility layer intrinsically has low value add, vendor concentration in private clouds will probably preserve prices and margins to some degree.

Leveraging network effects to deliver end to end application service management

In order to be able to deliver end to end online services, the upward-facing cloud software layer has to orchestrate and manage an untold number of application components, many from third parties, some from corporate developers. VMware has some leading-edge management technology that automatically wraps around applications (courtesy of the B-hive acquisition). But as long as commercial and corporate developers primarily target Windows as their application deployment platform, Windows will have a self-reinforcing advantage relative to VMware.

Microsoft’s self-reinforcing advantages are twofold, building on its leadership both as a deployment and development platform.

  1. First, since it accounts for roughly 80% of X86 server unit shipments, software developers of just about any stripe have to do at least some work to wrap Microsoft management tools around their applications.
  2. Second, because Microsoft is the leading provider of development tools on Windows, it will be able to capture even more management information about the subset of applications that use their tools. Even if both of these structural advantages haven’t yet been exploited by Microsoft, it’s hard to see that situation lasting as they roll out their Dynamic Systems Initiative.

VMware’s best potential for upside in this market is threefold.

  1. First, the extent to which commercial or corporate software developers use development platforms that are independent of Windows likely means neither company has an advantage in collecting management information. Examples of these platforms include J2EE, PHP, Ruby On Rails, Spring, and Hibernate. Some of these platforms don’t require any conventional operating system.
  2. Second, the extent to which VMware proliferates its platform ubiquitously, it may have more of an advantage in managing the infrastructure it sits on and the applications that sit on it.
  3. Third, virtualized environments make it possible to deploy applications in “appliances” that include all the bits required to run, including operating system, middleware, and multiple application components. Today, all these appliances are deployed using Linux because of Windows licensing restrictions. If appliances take off independent of Windows, that would help VMware tilt the platform competition in its favor.

Public clouds are likely to be more diverse, like a set of services, but with the core or anchor services having similar economics

Windows Azure has all the economic characteristics of a private cloud – masking the infrastructure, managing the application services – but with the margin-depressing overhead of tightly-integrated data centers. It will clearly have higher-level services like SQL Services, Exchange, and .NET, but it will be easy to integrate premise-based software as well as third-party services such as’s or Google Adwords.  In other words, if Microsoft delivers on its promise, it will be an anchor platform at a higher value, higher margin layer than Amazon, but with bridges to other services.  Considering how it appears to build on Microsoft’s on-premise software tools and interfaces, it appears likely to have a leading market share.

How VMWare And Cisco Might Bring The Nexgen Data Center Closer

September 14, 2008

Rumors of an alliance between Cisco and VMware have been swirling with varying levels of intensity for some time.

What’s the business problem that a potential alliance needs to address?

1) The most obvious one is that current VMware Distributed Resource Scheduling (DRS), Disaster Recovery (DR), and High Availability (HA) functionality built on core capabilities like VMotion are incomplete. It’s hard to move the Virtual Machine (VM) for spare capacity or to deal with downtime to any random server and maintain the connections to the same isolated data and storage area network (SAN). Instead, administrators either have to open up the network so any server can see any other server and any storage device, a security risk, or they have to manually remap the connections.

2) The less obvious and more speculative problem to be addressed is the management and automation of business services across resources and applications. It is still primitive, though the big 4, CA, BMC, HP Openview, and IBM Tivoli are all hard at work addressing this, and CIOs are looking for the provider of a strategic, new provider.


Economic Fallout From Virtualization In The Data Center

September 1, 2008

This is our first set of hypotheses about how virtualization is impacting each of the layers of the IT stack. We will elaborate and refine them as we continue to collect insights from vendors and our upcoming survey of IT decision makers.

The Ultimate Objective

· It’s more than just the savings from server consolidation and more than just greater flexibility in managing planned (VMotion) and unplanned downtime (disaster recovery, high availability)

· Ultimately, it’s about automating the data center in order to make it easier for companies to deliver online business and consumer services. The iconic example of an online service that complemented a traditional business was the Sabre travel reservation system born in the ‘60s. It was based on purpose-built infrastructure that required intense collaboration between the customer, American Airlines, and the vendor, IBM. More recent examples include Fedex package tracking or the familiar services from Amazon, eBay, and Google. In order to make it easier for businesses to build or assemble end to end services from existing assets, technology vendors have to convert “assets” into “pools of services” using virtualization at every layer of the IT stack.

Looking at the IT Stack Layer by Layer


When, If Ever, Will SaaS Crack Core, Mission Critical Processes In The Enterprise

July 17, 2008

It’s no secret Software as a Service (SaaS) has generated tremendous excitement among many customers for its apparently transformational adoption model and ownership experience.  Unlike client-server applications, SaaS delivers faster time to value often via a viral buying cycle as well as lower risk deployment.  The early adopter focus has been on small and midsize businesses (SMBs) because SaaS makes it economical to reach them with broad penetration for the first time.  Where SaaS has carved out successes in large enterprises, it has largely been in more independent, non-mission critical departmental functions who have no capex budgets such as HR, CRM, or marketing, not end to end suites.  Despite the undoubted progress that SaaS is making, we believe the adoption of core, mission critical processes (Financials, order management, industry-specific processes such as manufacturing or securities processing) in large enterprises is still many years out for a variety of technical and business challenges.

SMBs have been the early SaaS suite adopters because traditional vendors couldn”t reach them

SMBs have been the low hanging fruit for early SaaS adoption because they’ve historically been underserved by application vendors.  Small deal sizes and bare bones cost of ownership requirements typically were critical stumbling blocks.  The small deal sizes mean vendors have to reach them with a much lower cost channel than direct sales.


Why Selling Search To Microsoft Would Damage Yahoo’s Core Display Ad Business

July 7, 2008

I was really intrigued by David Kirkpatrick’s last story on Microsoft and Yahoo in Fortune Magazine.
His key insight was about search market share creating a more liquid marketplace for advertisers.   Search market share drives up the demand for and value of keywords.  That’s why Google’s search ad revenue share is even greater than its search query market share.

But there’s one other thing that Microsoft doesn’t mention when it talks about purchasing just “search”: there really is synergy between search ads and display ads, which is where Yahoo gets the majority of its revenues.  It’s very hard to measure the value of display ads because they don’t surface when a user does something that shows their immediate intent (e.g. search for a lawnmower) – that’s why they can only charge for eyeballs/impressions while search engines can charge for clicks on keywords.

But everyone knows that those display ads create brand equity or other awareness that search ads ultimately get to monetize.  So all the major online vendors (MSFT, Yahoo, GOOG) who happen to have both major display ad networks (Google’s I think was built on Doubleclick) and search engines are planning to track what display ads users look at over time.  Then they will be able to correlate that with later search queries and search ad click-throughs.  In other words, they are finally going to start attributing more value to the display ads, which have never really been measurable.  But Microsoft has a plenty big display ad network so they just want to carve the search business from Yahoo.  And Yahoo is under such pressure to do a deal that they’re willing to sacrifice the future synergy of their search and display ad businesses.  Without search, they won’t be able to offer advertisers the same measurement capabilities as Microsoft and Google.  Their display ad pricing and revenues will suffer as a result.

They have been unable to articulate this synergy, or any other part of their strategy for that matter. It’s ironic, but they probably would be able to leverage that synergy with the Google search deal.  They also seem to be turning their internal platform for their properties into an Internet scale Web developer platform.  But they can’t even convince their top talent to give them some breathing room.  It’s game over.

When Applications Talk To Each Other Via SOA, What Happens To User-Based Pricing

June 16, 2008

Alphabet soup of evolving application design patterns: SOA, EDA, BPM

It’s clear that SaaS doesn’t represent a threat to client/server pricing.  Consider what SOA might represent. In case the terminology is new, here are the definitions first. Bear with the abstractions. To be technically correct, we have to include Event-Driven Architecture (EDA) and Business Process Management (BPM) technologies as well for customers to get the full value out of autonomous services communicating with each other with few users involved.

  • In the case of enterprise applications, SOA means functionality in the form of business processes that are made up of services that communicate with each other’s interfaces by exchanging data. These interfaces might be implemented as Web services. The supplier electronically submitting an invoice to a customer is the relevant example here. 
  • EDA allows services in an SOA to be more loosely coupled. They can communicate by publishing and subscribing to events without either side talking directly to the other. A retailer tracking delivery of goods to its distribution center via RFID would be an example here.
  • BPM orchestrates services and includes users in a workflow where necessary to manage a complete process. A BPM agent for a supplier may be managing the order to cash process when a customer places an order that goes beyond their credit limit. The BPM agent may escalate this exception to the finance department as well as the sales account manager for resolution.


Why SaaS Isn’t The Real Threat To Enterprise Application Pricing

June 16, 2008

Whether subscriptions or perpetual licenses, it’s still about user-based pricing

Imagine for a moment that you are at IBM and a small supplier of components from the Far East has just submitted an invoice. It just shipped an order of printed circuit boards to IBM’s networking equipment division in upstate New York. IBM receives it and a clerk in its invoice processing department enters the invoice into its ERP system. Whether IBM bought a client/server or software-as-a-service (SaaS) ERP system doesn’t matter. The clerk has to fill out and navigate as many as 20 screens to enter the invoice so the purchase to payment process can move to the next step.

But go back to the distinction between client/server and SaaS applications. Conventional wisdom says that SaaS applications such as and their subscription licenses represent a threat to the perpetual licenses and the business models of traditional client-server companies such as Oracle or SAP. Stretching payments out over multiple years, as SaaS does, makes it harder to show the profitability and growth that comes from the upfront payments of perpetual licenses. The reality is somewhat different. As many know, SaaS actually takes in significantly more revenue over the product’s lifecycle. But the pricing models have much more in common than their differences. They both charge based on the number of users accessing the application.


Making Advertising Work On The Other (Non-Search) Part Of The Web

June 3, 2008


Just about everyone knows advertising on the Web, outside of search, isn’t living up to expectations. Advertisers and publishers buy and sell inventory of display ads for CPMs (cost per thousand impressions) well below the offline equivalents in magazines, newspapers, and tv. Part of the reason is that outside of search, the Web hasn’t found its equivalent of the 30 second spot or the two page spread (an insight courtesy of John Batelle). But a major part of the reason is that Web sites can’t consistently serve up relevant and engaging ads to an audience that’s no longer captive. So how do we substitute relevance for the more traditional reach to increase ad prices?

For years users have left “breadcrumbs” about their interests across the Web. Now there are increasingly sophisticated ways to connect that information into rich profiles, much of it by user choice, while still respecting privacy concerns. The consolidation and massive reach of ad networks in the hands of Google, Yahoo, and Microsoft likely means that most of the economic rewards of this shift will continue to accrue to these companies. Even if the traditional media giants accelerate their migration to the Web, they are likely too late. They won’t be able to match the reach of the tech giants’ online ad networks and their trove of online personal profiles.


The Possible Paths From Today’s Virtualization To Cloud Computing

May 26, 2008

George Gilbert

From Virtualization To Cloud Computing

Virtualization and cloud computing have been getting a ton of buzz. But there has been less discussion about how virtualization, now known mainly for its server consolidation capability, will morph into cloud computing. For that to happen, servers, storage, and networks have to all fuse into one virtual machine from a developer’s and an administrator’s perspective. If the rumors that Cisco will buy EMC (and by extension its majority stake in VMware), the industry will have the first vendor who has a credible shot at putting together all the pieces. This post and the one that follows attempt to layout the different ways this transition could unfold. (Disclosure: I’m an investor in VMware).

Cloud computing, previously known as utility computing, is where all computing resources in Internet data centers look to users, developers, and administrators like one giant computer. It offers seamless scalability and radically reduced administrative overhead. There is more than one path from today’s virtualization to tomorrow’s cloud computing, and they’re not necessarily straightforward.

Ray Ozzie highlighted the importance of the transition from virtualization to cloud computing as one of the “three core principles that we’re using to drive the reconceptualization of our software so as to embrace this world of services that we live in… Most major enterprises are, today, in the early stages of what will be a very, very significant transition from the use of dedicated application servers to the use of virtualization and commodity hardware for consolidating apps on computing grids and storage grids within their data center. This trend will accelerate as apps are progressively refactored, horizontally refactored, to make use of this new virtualization-powered utility computing model. A model that will span from the enterprise data center, and ultimately, into the cloud…”


Elaborating On The Scenarios From Virtualization To Cloud Computing

May 25, 2008

George Gilbert

In the last post, I outlined why virtualization was morphing into cloud computing.  In this post, I elaborate on the potential paths it could take.

1.  VMware manages compute virtualization, Cisco manages network virtualization, and another vendor such as EMC or Network Appliance manages storage virtualization:

In this scenario, VMware provides the developer and management interfaces for making all the servers look like a single machine.  But customers adopt Cisco, which recently introduced its Nexus line of switches, as the network virtualization layer.  This product creates virtual networks and connections between computers and storage networks out of a physical switch. There are a variety of approaches to storage virtualization, but for the sake of simplicity let’s say companies choose to deploy EMC or NetApp.  Again, developers and administrators see only one logical device.  The downside of this approach relative to one vendor owning all the virtual resources end to end is twofold.  Software developers have to write to three separate interfaces to work with the cloud.  Second, administrators also have to work with three consoles to make sure software can deliver on its SLAs.

2. VMware becomes the end to end infrastructure:

Today VMware only offers virtual compute infrastructure.  It would still need to offer file system virtualization and network virtualization.  And of course, it would have to build the whole policy-based management infrastructure, or at least a framework other vendors could plug into to complete the platform.  The challenge with this scenario is that VMware has the reputation of being somewhat closed.  So the burden of the storage and network virtualization work, which is non-trivial, would fall mostly on VMware.

3. Microsoft manages end to end virtual and physical resources for Windows shops:

Microsoft has made a lot of noise with its Hyper-V server virtualization product and the emerging suite of management tools it is promising that go along with its management tools for physical resources.  Although it also supports SUSE and Red Hat Linux, it’s possible it could spread its footprint in the Windows environment to support storage and network virtualization with the proper hardware partnerships.  For Windows-only shops, managing all the physical and virtual resources with one set of interfaces for developers and administrators would be ideal.  I don’t know how difficult it would be to accomplish the storage and networking portions.

4. Red Hat or Citrix could do the same as 1 or 2 for Red Hat Linux shops

Although XenServer supports more Linux distributions than Red Hat, plus Solaris on X86, it’s hard to see application developers and system administrators committing to another distribution for end-to-end deployment.  The challenge with a Red Hat deployment as platform for end to end virtualization is that they’ve lost control of their virtualization technology to Citrix.  And it’s hard to see application developers committing to API’s promoted by a firm known for terminal services.

5.  HP Openview or IBM Tivoli manage both virtual and physical infrastructure in mulit-vendor shops:

At the beginning of the decade, this was the default assumption held by industry analysts and probably most customers.  The core assumption for this scenario today is that these are the incumbent vendors for multi-vendor shops and they are the only ones who can bring order to the chaos.  The challenge they face is that they appear to have almost no presence in the market for virtualization infrastructure right now.  Their server businesses are among VMware’s biggest partners.  They are in no position to define the developer interfaces to virtualization products.  A more likely scenario is that they integrate their management tools with VMware, Xen, and Microsoft’s Hyper-V.

6.  Individual vendors like Oracle and SAP write their own policy-based virtualization into their infrastructure:

Oracle already announced support for the Xen server but SAP’s plans for multi-tenancy are less clear.  But in this scenario, each vendor builds virtualization support into its own products.  In Oracle’s case, presumably this would be the database, application server, and the business applications that support them.  In SAP’s case, this would probably mean its NetWeaver application server and the related business applications.  This vertically integrated approach has some challenges of its own.  Customers running SAP on top of Oracle infrastructure, for example, would have conflicting policy-based administration layers trying to ensure SLAs.

7. Cloud computing vendors create their own purpose-built virtualization infrastructure

Rather than adopt the commercially available products, the major cloud computing vendors such as Google, Microsoft, Yahoo, Amazon, and, etc. all build technology specifically for their platforms.  So far, all but Microsoft seem to be pursuing this path.  However, only Microsoft is talking specifically about seamless scalability from running software hosted on the customer’s premise to cloud-based deployment.  If customers are interested in seamlessly migrating their enterprise software into the cloud, other than Microsoft, they’re going to have trouble working with these vendors.  More likely, these vendors are going to be the platforms for a new class of consumer-facing Web software.  Microsoft and other yet to emerge vendors are likely to be the cloud platforms of choice as today’s enterprise software migrates to the cloud.