Digital Walkabout

Microsoft Live Mesh and the Cloud

Microsoft's Live Mesh has been recognised by the software vendor as a ground-breaking Cloud technology and so has been elevated in their strategy, part of which is developing an ecosystem around Live Mesh applications. One of the first deliverables to this end will be a set of applications that build on top of Live Mesh, using the Live Mesh SDK. 

“The Live Mesh cloud services and client platform provide powerful FeedSync-based data synchronization capabilities, device P2P and cloud-relay communications, pub-sub infrastructure, and an extensibility model for applications. This session describes how you can take advantage of the Mesh developer platform, protocols, and APIs to mesh-enable your existing and future web services and client applications-allowing you to target unique new scenarios and reach new users.”

Another pivot is the Mesh Operating Environment, or MOE, “a service composition runtime that provides a unified view of all Live Mesh services and a simple way for applications to interact with Live Mesh.  MOE is everywhere—it’s on all devices in your mesh (as ‘client MOE’), and it’s in the cloud (’cloud MOE’).”

Thirdly, Live Mesh, together with Silverlight, provides the runtime on which Mesh apps will execute. The demo Live Mesh “Tracker” application highlghts this, executing from either the Mesh desktop in the browser or on the Windows desktop as a real application. This is achieved by the Mesh apps running inside chromeless Internet Explorer and Silverlight wrappers provided by the Mesh Operating Environment. 

Live Mesh is certainly a space to watch, as it has the potential to enable a new class of Cloud applications. The P2P architecture promises much in the way of scalable Cloud applications leveraging all involved resources, which promotes reduced OPEX for the data centre. 

The genius behind Google Chrome

FT has published a (non technical) interview with Lars Bak on his v8 work in Chrome. It highlights the importance of the browser as the main cloud client and within browsers, the importance of the Web 2.0 (JavaScript) runtime. An extract:

<<The sophistication of web applications is increasing much faster than the corresponding power of the browsers. It’s like having lots of high-performance sports cars using rutted, bumpy roads. And yet no car manufacturer has ever invested in road building. That sort of move would, after all, benefit competitors equally, not to mention being massively expensive. Software development doesn’t cost as much but the competitor problem remains. But Google says it doesn’t mind: without an improvement in browser performance, the company argues, we all suffer.

The ability to access complex programs through the web is known as “cloud computing”, and Google isn’t the only one arguing that it’s the way ahead. Even Microsoft, the company synonymous with applications and operating systems located on the desktop, has talked about moving into “the cloud”. Steve Ballmer, Microsoft’s chief executive, has promised an “operating system that runs in the internet” – something he has dubbed “Windows Cloud”. But for cloud computing truly to take off, web browsers need to be better.>>

Comparing the Cloud Platforms

Since the announcement of Cloud Computing platforms such as Amazon Web Services, Microsoft Windows Azure, Google AppEngine and VMware vCloud, there has been a swell of interest in what they provide, how they compare to one another, and which should those developing and deploying Cloud applications should consider.

In order to do this, it is important to understand the principal Cloud models, since these map to the various scenarios which apply to both consumer and enterprise applications. Here is a summary of the models:

Hosted Services: Thin client, centralised architecture, data and capabilities exclusively in the cloud

Attached Services: Traditional desktop client, client-server architecture, data and extended capabilities in the cloud

Rich Internet Services: Rich internet client, p2p architecture, data and capabilities distributed, clients part of the cloud

Here is a summary of the four platfoms:

Microsoft Windows Azure: Currently in private beta; boasts an impressive development platform, integrated set of capabilities, an strong Cloud architecture (supporting Hosted Services, Attached Services and Rich Internet Services) and the promise of robust enterprise support and integration. Presently the beta provides public Cloud capabilities through both PaaS and IaaS and operates on the principle of a Cloud OS rather than virtualised environments, but the IaaS capabilities are entirely opaque to the user, being managed by Microsoft. The private Cloud version still in development would enable IT to host their own Windows Azure Cloud. All development, deployment, administration, management and monitoring are provided at the PaaS level.

Amazon Web Services: The first solution to exit beta, AWS provides the most mature public Cloud solution on the market. AWS offers IaaS, - the compute resources for virtual machines, storage, messaging, persistence - operating on the principle of virtualised environments, affording the user full control over virtual machines and to deploy and operate LAMP- or Microsoft-stack applications to their requirements and preferences. For PaaS, the user has to use a third party platform or assemble their own. As of this date there are no plans to provide a private Cloud version of AWS.

Google App Engine: In beta, Google AppEngine claims that one can "drop your code and we’ll figure out the rest”. AppEngine provides SaaS with carefully constrained PaaS, enabling users to build their apps and deploy them to the Google's environment, where they manage all of IaaS and most of PaaS. Google's architecture supports only public Cloud,, and as of this date there are no plans to provide a private Cloud version or analogue. AppEngine, like Google Apps, only supports the Hosted Services model.

VMware vCloud: Announced in late 2008, vCloud is solution promising to enable taking standard VMware virtual appliances and runing them anywhere: on-premise or in a data centre of any provider supporting VMware’s infrastructure. Similar to AWS, vCloud provides IaaS, the thrust of VMware's proposition being that it is to support both public and private Cloud on release. This differs from the other four in this comparison; only Microsoft has indicated upcoming support for private Cloud, and this trails its public Cloud beta.

A breakdown of these four platforms by features and characteristics is as follows:

Features/ Charateristics	Microsoft	Amazon	Google	VMware
Availability	Private CTP beta	General availability	Public beta	Announced
Supported Cloud Models	Hosted Services, Attached Services and Rich Internet Services, through the PaaS layer.	Hosted Services, Attached Services and Rich Internet Services, through the VMs.	Hosted Services	Hosted Services, Attached Services and Rich Internet Services, through the VMs.
Computing Architecture	Open PaaS + Closed IaaS: Develop apps (.NET) for clients and servers, the latter which Microsoft runs on Windows 2008 VMs according to environment specifications (number of machines and types needed, etcetera).	Open IaaS: Elastic Compute Cloud (EC2) supports uploads XEN virtual machine images to the infrastructure and provides services to instantiate and manage them.	Closed SaaS + Closed Paas: Develop web applications in Python or Django within specific parameters, deploy to AppEngine.	Open IaaS: Enables one to easily move virtual machines between environments and run them onpremise or at any partner data centre.
Load balancing	Yes	Yes	Yes	Not announced
Storage	App and file storage; SQL services	Simple Storage Service (S3) and SimpleDB	Datastore APIs	Not announced
Message queuing for machine communcations	Message queue in Windows Azure storage	Simple Queue Service (SQS)	None	Not announced
Integration with other services	.NET services as BPM integration(i.e. BizTalk in the cloud): Identity management, access control , and workflow services, as well as a service bus. Live services (contacts, mail, maps, etcetera)	None	Integration supported with existing Google services: authentication, mail, base, calendar, contacts, documents, pictures, spreadsheets, YouTube.	None
Public or Private Cloud	Public, private expected in late 2009	Public	Public	Public and private
Development tools	Integration into Visual Studio, support for any .NET languages, support for third party IDEs (starting with Eclipse). Support for other languages (Python, Java)planned for late 2009. Strong story for development as this is supported by the PaaS layer.	Not applicable. Amazon runs VMs, and as such does not provide any development platforms. One develops apps as one otherwise would for the base OS running on VMs. No PaaS support for development.	Basic editing, local simulation, and deployment tools. Language selection limited to Python and Django. Application-level tools such as Google Web Toolkit (GWT) do not have any integration with Google AppEngine. Limited PaaS support for development.	Not applicable. VMware runs VMs, and as such does not provide any development platforms. One develops apps as one otherwise would for the base OS running on VMs. No PaaS support for development.

Which is best for the user and IT? This depends on the type of applications one plans to deploy to the Cloud, a rich topic in itself, which I leave to my presentation "There's Cloud, then there's Cloud".

Fog alert: Cloud Computing FUD

McKinsey & Co recently released an analysis which purports that Cloud Computing does not make sense for the enterprise, based on assumptions regarding a) CC being represented only by off-premises hosting and thin client architectures, and b) the true TCO of the data centre. In a considerable body of work by many authors, including myself, we have endeavoured to raise the awareness of what is actually represented by Cloud, the multiple models and how these fit various scenarios, from the entry-level user to the enterprise. If it seems unclear as to why we collectively carry on vociferously, the McKinsey analysis should erase any such confusion. Fear, uncertainty and doubt (FUD) are the greatest threats to both technology as a whole and the benefits and value-generation gained from technology adoption. FUD transforms information into mis-information faster than a game of Whispers, and serves only to promote rumour above fact. 

The study uses Amazon.com’s Web service offering as the price of outsourced cloud computing, since its service is the best-known, and its costs are published. On that basis, McKinsey asserts the total cost of hired data centre functions would be $366/month per unit of output, compared with purchased resources of $150/month for the conventional data centre. “The industry has assumed the financial benefits of cloud computing and, in our view, that’s a faulty assumption,” said Will Forrest, a principal at McKinsey, who led the study.

I have found three principal issues with McKinsey's analysis:

1) Without firstly recognising that CC encompasses more than Hosted Services, McKinsey is in no position to make any of sweeping statements about "Cloud Computing". The swift and pointed reaction to McKinsey's analysis makes this quite clear. Until McKinsey spends cycles better understanding the fundamentals, it is what seems not to make sense, rather than CC. There is data in McKinsey's analysis which is rather useful for evaluating a move to CC, but this is only occluded by the egregious assumptions and the questionable conclusion drawn from them. 

2) Even Hosted Services continues to lower costs as economies of scale develop. The Hosted Services model of CC converts this to a mass infrastructure, from the present scenario where software is licensed on a cyclical basis. Now, to go "unto the breach once more", it has become only painfully clear that Hosted Services is only *one* model of CC. Salesforce.com and many other organisations developing and/or deploying Cloud applications and platforms have, in no uncertain sense, illustrated the progression toward the Rich Internet Cloud and the value that has generated. Point being: CC is not monolithic, and amongst the models are found economies of scale, lower TCO, greener IT, and more reliable compute models. Interestingly enough, hardware, software and services vendors all have tremendous opportunities to benefit from and excel in CC. The only organisations who stand to risk losing business are traditional management and IT consultancy, and IT outsourcing intermediaries. All three CC models help cut through the bureaucracy to better support globalised virtual teams, which reduces the need for this sort of intermediation and consultancy. 

3. McKinsey's data assume the total elimination of hardware acquisition and replacement with virtual Hosted Services resources (from Amazon Web Services). This is not the business case for Hosted Services CC, much less for Attached Services or Rich Internet Cloud, and serves only an "apples-to-oranges" comparison. Among the principal business cases of CC is to dynamically scale up resources during peak demand, whether the public Cloud/Hosted Services model of supplementing locally operated private Cloud with outsourced resources, or the private Cloud/Rich Internet model of shifting cycles amongst the private Cloud of client and server resources to adapt to demand.

The valid comparison would be the practice of buying incrementally to use during peak periods, as opposed to hiring off-premises resources for such intervals. The most basic analyses, utilising McKinsey's own data, yield actionable information, which only further support the Cloud's value proposition of dynamic resourcing. Add to this the analyses of Gartner (http://bit.ly/NT4zv http://bit.ly/11IItA) and IDC (http://bit.ly/13M6uJ), and it becomes rather difficult to challenge the value proposition.

The McKinsey analysis raises many questions as to its motivations - whether McKinsey is trying to throttle a paradigm that could make organisations across the globe more competitive; or simply failing to understand what Cloud represents and its implications, and falling victim to FUD. My opinion is that a bit of both have happened. The attempt to position virtualisation as a competing paradigm, rather than a convergent one, clearly reflects a lack of understanding of this space, as does the egregious representation of Cloud Computing as a monolithic model. 

Yet, as has been made painfully clear in all aspects, there is no such thing as bad publicity. :)

Thumbnail sketch of CC economies of scale

Based on the McKinsey analysis and assumptions of the total cost of the data centre functions hired from Amazon Web Services would be $366/month per unit of output, compared with $150/month for the conventional data centre, here is a rough comparison of the practice of buying incrementally to use during peak periods, as opposed to hiring off-premises resources for such intervals.

The purpose of this thumbnail sketch is *not* to argue for or against purchasing resources for private Cloud or hiring resources from AWS. On the contrary, it is to illustrate how McKinsey's own data can easily illustrate an entirely different conclusion.

Short-term hire: Additional resources are required for a peak of 2 months duration. 
Purchase: $150 *12 = $1,800 per year per unit of resources, of which the utilization is 2/12 = 16% 
Hire from AWS: $366 for 2 months = $732/year per unit assuming 100% utilisation

Value: $1,800 - $732 = $1,068/year, or 59% gain from hiring AWS CC resources. 

Long-term hire: Additional resources are required for a peak of 9 months duration. 
Purchase: $150 *12 = $1,800 per year per unit of resources, of which the utilization is 9/12 = 75% 
Hire from Amazon.com: $366 for 9 months = $3294/year per unit assuming 100% utilisation 
Value: $3,294 - $1,800 = $1,494/year, or 83% gain from purchasing resources for private Cloud. 

Thus, using McKinsey cost data and assumptions, one may conclude that it makes sense to hire AWS resources for a maximum of 5 months/year; beyond that, one gains more from purchasing resources for one's own private cloud. 

One more observation: I highly recommend reading Darrel Huff's "How to Lie with Statistics". 

BarCamp Transparency

Thank to Amanda Clark, a Barcamp Transparency is being organised to talk about social media ethics, Web 2.0 ethics, Open Government, cyber activism, direct democracy and all things related to transparency. 

http://barcamptransparency.pbwiki.com/FrontPage

Making the case for Private Cloud

Reviewing the value propositions of today's principal (at least in mindshare) Cloud Computing environments, Microsoft Azure/Pangea, IBM BlueCloud and Amazon EC2/S3, some of the strategies began to make sense to me. Amazon provides full control over a virtual environment such as a Windows Server 2008 or Ubuntu instance - much like a Virtual Machine - whilst Microsoft requires coding their API to take advantage of their architecture. An initial reaction to the Azure implementation by many questions why one would specifically build an application to fit Microsoft’s Cloud - a VM could just as easily be created and uploaded to AWS or a hosted Microsoft Hyper-V environment. 

Reading the signals, it is apparent that Microsoft is looking to private Cloud. Enterprise IT continue to harbour reservations to hosting applications in the public Cloud, and in my experiences with various organisations, I have encountered this sort of resistance on the basis of security, isolation, privacy and management concerns. Without robust solutions for these gaps, multi-tenant public Cloud hosted services have the potential for competitors to have access to one another's corporate data; the concerns over the compromise of Sarbanes-Oxley regulations are greater. 

Thus, the appeal of private Cloud. What could be Microsoft’s strategy here? One can host a private Cloud with a cluster of Windows 2008 Servers, SQL Server database instances, and Windows Azure. To deploy additional application resources, there is no need to spin up a new VM or physical host requiring naming, updating, configuring, provisioning, or managing.  IT personnel or automata simply scale  up the application, and license management for these resources is addressed at the cluster level. 

As to licensing based on CPU usage, this provides additional scalability - when apps are processing a heavy load, say between 8h and 18h, they automatically scale out, and during off-hours, they are reduced. Batch and data store processes that execute during off-hours are then scaled out. Thinking through the scenario, this has significant implications for recovery management, power management, resource management, and more. Full utilisation of the data centre moves closer to reality, and with Infrastructure completely abstracted for applications, building this a Microsoft Cloud platform begins to make sense. 

Microsoft has given mixed signals about this. Their currently position Azure as another public Cloud play whilst indicating that a private Cloud product is on its way. Pangea is Microsoft's more ostensible private Cloud play, yet Microsoft has yet to go to press with any indication that its private Cloud strategy is anything more than platform hooks and bundled offerings provided with Windows 2008 Server. Given the existing demand and momentum for private Cloud, not to mention the potential of this market, Microsoft has an opportunity to make tremendous impact here. Clearer and less reactive positioning and messaging around this from Microsoft would only increase the multiplier.

Do we really need a Cloud Manifesto?

Reading this InformationWeek article, I thought over this topic again. The short of it: I'm also not so certain that we really need a Cloud Manifesto. Rather, I think we need a reference metamodel to clarify what is Cloud and a common Cloud taxonomy, and associate this with reference architectures. There is considerable FUD around what constitutes Cloud, which is something I have addressed at length (e.g. Scalable Cloud Computing), to which a Cloud-RM would go far to address. Sincerely, I believe we are at an early stage for this - many vendors, including Intel, Microsoft, IBM, Amazon, SalesForce.com, Greenmonk.net, and analysts have presented extensively on this. To some degree or another, and whether using the terminology I use or not, these have all spoken to the three principal compute models of Cloud: 

• Hosted Services (thin client, centralised architecture, data and capabilities exclusively in the cloud)
• Attached Services (traditional desktop client, client-server architecture, data and extended capabilities in the cloud)
• Rich Internet Services (rich client, p2p architecture, data and capabilities distributed, clients part of the cloud). 

As this RM and taxonomy continue to develop, their mapping to various platforms and architectures is a considerable aide to CC entrants and consumers (e.g. Web 1.0 is an analogy of the Hosted Service model, whilst Web 2.0/3.0 can be compared to the Rich Internet Service model; likewise, look at the evolution of SalesForce.com, from Attached Services, to Hosted Services, to Rich Internet Services to understand the logic of such progression).  

I see the value in a reference metamodel that does this - guiding users, deployers and developers of CC so as to understand their requirements, positioning and paths. I believe that the SOA-RM, of which I was a part, had the right idea in mind. With the level of vendor and developer momentum and support for CC, a reference metamodel is achievable. Do we really need another standard? Or can we have those who want to clarify positioning and messaging propose a metamodel and taxonomy based on the organic convergence of these things in the CC world? Only time will tell. 

The CC Value Proposition for Utilisation

Averaged over a business year, most servers have a CPU and network bandwidth utilisation of 1% or less. Storage capacity on disc that can be accessed only from a specific servers are also under-utilised - e.g. disc capacity attached to a database server, used only when views, some queries, and other interactions which require RDBMS temporary constructs store intermediate results to disc. Isolated pools of computing, network and storage are under-utilised most of the time, but must be provisioned for peak capacity, whether that is measured by day or hour. 

As such, it is not surprising that publicly available research supports an average utilisation of 10-15%. If one extrapolates this, theoretically 85% to 90% of installed computing capacity is sitting idle. Utility computing has improved this utilisation rate dramatically - HPC packages such as IBM BlueGrid processing can increase utilisation close to 90% - yet adoption has been limited. 

CC is the latest evolution and convergence of utility compute - together with web compute, and leveraging virtualisation capabilities - and as such, one can only expect it to increase utilisation rates. Yet there are several considerations that need to be taken before going off on a wild goose chase.

There are the technical considerations of HPC modes - which requires exclusive host access for certain processing. Next, once this is metered, one can reserve compute resources in advance, which will tip the focus from maximising utilsation to maximising revenue. One will not place as much import on low utilisation, given that full capacity is being reserved and billed. CC providers thus will rather take their concern to the idle percentage of un-reserved compute resources. If one thinks about this very carefully, they will see the strong similarity to hotel and air carrier capacity management, and other logisitical problems. 

Existing software vendors with per Server and per CPU licensing models have much to lose from this shift. A BI/BPO provider such as Vitria, provisioned across the Cloud, can compete very well with traditional datawarehousing frameworks. One can use their tools for only part of the year, yet consume multiyear data, using temporarily provisioned servers and on-demand software. The TCO of this approach is easily an order of magnitude less than what can be referred to as "Traditional IT" (i.e. the buy, install, maintain and manage model). 

With respect to licensing, it is interesting to note how Microsoft has been influenced to license MSOffice by real-time use rather than per seat installation (c.f. MSOnline), with WalMart and other consumer channels leading the charge. AT&T U-verse has likewise influenced Microsoft to license MediaRoom by real-time use. Does this mean WalMart will sell TVs,, or AT&T sell TV programming, on a per-use basis?

I expect CC vendors' private cloud offerings will provide the tipping point that will persuade mainstream IT to adopt this approach. With a private cloud, one could share compute, network and storage resources amongst departments, business units, subsidiaries, etcetera. With the ready-made showcase of consumer services, which themselves represent public enterprise computing (c.f. Windows Live), this isn't far away. Take a look at this insightful comparison of existing CC offerings for further rumination. 

Amazon Elastic MapReduce

AAmazon is beginning to move from being a pure IaaS play to supporting more abstractions - here are some details w/r/t the Hadoop service from the AWS blog:

Today we are rolling out Amazon Elastic MapReduce. Using Elastic MapReduce, you can create, run, monitor, and control Hadoop jobs with point-and-click ease. You don't have to go out and buys scads of hardware. You don't have to rack it, network it, or administer it. You don't have to worry about running out of resources or sharing them with other members of your organization. You don't have to monitor it, tune it, or spend time upgrading the system or application software on it. You can run world-scale jobs anytime you would like, while remaining focused on your results. Note that I said jobs (plural), not job. Subject to the number of EC2 instances you are allowed to run, you can start up any number of MapReduce jobs in parallel. You can always request an additional allocation of EC2 instances here.

Processing in Elastic MapReduce is centered around the concept of a Job Flow. Each Job Flow can contain one or more Steps. Each step inhales a bunch of data from Amazon S3, distributes it to a specified number of EC2 instances running Hadoop (spinning up the instances if necessary), does all of the work, and then writes the results back to S3. Each step must reference application- specific "mapper" and/or "reducer" code (Java JARs or scripting code for use via the Streaming model). We've also included the Aggregate Package with built-in support for a number of common operations such as Sum, Min, Max, Histogram, and Count. You can get a lot done before you even start to write code!

We're providing three distinct access routes to Elastic MapReduce. You have complete control via the Elastic MapReduce API, you can use the Elastic MapReduce command-line tools, or you can go all point-and-click with the Elastic MapReduce tab within the AWS Management Console! Let's take a look at each one.

 

NYT published an article yesterday on the increasing adoption of Hadoop beyond search. One simple way in which our NM technology can be combined with Hadoop and a Ganglia power monitoring plugin can be seen in the foil extract below.
 
It is also important to note that the technical ideas behind Hadoop/Map Reduce have roots in the functional programming world. Functional programming itself is expected to become increasingly important in the many core era (referential transparency/side effect free/Idempotent functions are important enablers for parallelization/distribution of computing).