Think Way Outside the box – Microsoft’s Azure into cloud computing

Finally Microsoft has entered into the cloud business formally with Windows Azure. These are my notes from MS PDC 2008.

Executive summary:

Windows Azure is not an OS but an offering. It consists of :

• A hosting environment to deploy your services (you define the rules & provide your code, the platform will take care of the rest!) for a spectrum of users – from hobbyists to enterprise developers
• Automated service management (abstracts hardware, load balancing and a host of other similar functions, based on the service model you create, which has things like service topology, size, health constraints and so on)
• Scalable storage
• A rich developer experience(This is where Microsoft has leverage- the Azure fits seamlessly into their development environment- you can write usual code, test it in their cloud simulation environment, debug the code and then deploy it to the cloud. So the current development skills are fully transferable! The deployment is so easy “even a CEO or a VP can do it!”)
• Windows Azure as a services layer with .NET services (service bus,access control and work flow services)
• They also offer SQL Services, SharePoint services and CRM on this platform

Their perspective of a cloud is very simple – “A set of connected servers;on which install and run services;and store and retrieve data” – and their offerings reflect that view of the world

Their view of the business requirements on a cloud infrastructure are:

• Interoperability and Business Processes
• Identity and Security
• Data management & Compliance
• Services Management

Some quick thoughts

• In terms of impact, Bob Muglia compares this year to PDC ‘92, when Microsoft announced Widows NT
• Microsoft characterizes this as 5th generation of computing
  • Monolithic(70s), Client Server(80s),web(90s),SOA(now)-services(2009+)
• They are not embracing the term Cloud, but are calling it services exposed via web protocols!
• Their motto is software + services -> The power of choice
• They see Azure as helping to evolve existing paradigms to work with hybrid architectures
• Also suddenly there are fabric everywhere – Azure service management fabric, development fabric,… <KS>Neiman Marcus anyone ?</KS>
• All features are not currently available, they are “exposing functionality in a staged manner”
• Introducing the cloud, they have a little arrogance and one would come away with the feeling that Microsoft had been at it for years ! Ray did tip his hat to Amazon and Jeff, though.
• From a business perspective, Microsoft s formally in the infrastructure business

Gory Details:

Day 1

Keynotes
Ray Ozzie introduced the concepts and Azure, followed by Amitabh Srivastava, Bob Muglia and David Thompson detailing different parts of the offering.

• Current applications have enterprise as the scope and with cloud the scope has expanded – Cloud is the externalization of IT
• The separate roles of software developer and operations are intertwined with the cloud computing paradigm
• While some companies have the resources required for the operation discipline to run a global infrastructure,many find it a disproportionate burden
• Some challenges to be solved by cloud computing
  • Meeting customer expectation of interactive, participatory web systems
  • Operating across peaks and valleys
  • Continuity Issues
  • Loosely coupled architecture, data replication strategies, data partition strategies
  • Ray calls clouds “overdraft protection for your web site”
• This “high scale internet infrastructure” is a new tier
  • Desktop Tier – scope of a single machine
  • Enterprise Systems – scope of the enterprise
  • Web Tier – scope of the web
• Windows Azure is not an OS but Kernel of cloud platform. “Kernels do not demo very well”, so they showed demo of a few apps

• Business View of Azure
  • Scalable infrastructure
  • Ability to manage a large global datacenter infrastructure
  • Federated DataCenter (This is one of their key themes)
  • Automated Service Management
  • Applications and OS managed separately
  • Fabric Controller manages lifecycle

• The service model is one of the key concepts
  • That is how one defines a service declaratively. The model includes roles.channels, adaptors, interfaces, configuration setting et al
  • The service model is an XML file
• The services bus is another important piece because it securely connects the on perm and cloud seamlessly even through firewalls
• The have a good Identity substrate with connectors and gateways to transcend between on perm and cloud. <KS>This I thought was a good value proposition; seamless identity across enterprise and cloud is difficult. I also saw that they now support openID.Good move</KS>

Odds & Ends

• An interesting URL!
• Azure white paper
• Yep, I know – Azure is not a verb. May be it should ;o)

What is cloud computing and do I need to be scared ?

This interesting question asked by the spouse of Tom Hogan(HP), as he was preparing for the keynote of the Cloud Summit Conference by SandHill (media coverage here, some of the presentations here) While the focus of the conference was not about clouds on Friday the 13th, the discussions were around separating hype from reality (unfortunately a lot of it is going on these days ;o() in a very business sense. As MR pointed out during the introduction, after hearing the talks and discussions at the conference, one can either celebrate or drown one’s sorrows, during the cocktail reception ! MR’s view quoted here sums up some of the points raised in the conference.

The conference started with MR’s introduction and Tom Hogan’s keynote followed.

Cloud Computing and the Enterprise

Summary of Tom’s talk:

• Requires shift to new architectural changes including multi tenancy, security and availability as applied to a cloud infrastructure(<KS> [Update: 10/22/08] Addition after Neil’s comments. It is not that security and availability are not part of current IT architecture, but that the mechanisms are slightly different in a cloud environment and architectures need to account for that</KS>)
• Business value propositions – deliver efficiency, increase speed and agility, mitigation of risk, enable alignment  & outcome based deployment of resources(<KS> [Update : 10/22/08] Addition after Neil’s comments. Naturally all these attributes are part of current IT infrastructures. But a cloud infrastructure has a definite advantage over traditional infrastructure, in terms of business experimentation, efficiency, peak scalability, agility and responsiveness</KS>)
• [Update : 10/22/08] Good responses from Neil and Kannan below]
• IT=BT (Business technology)
• Currently 65% of IT budget goes to operation, 25% application management and 10-15% on innovation. If cloud infrastructures can make a dent in the 65% that is a big advantage as the savings can be applied to innovation
• There is a chasm (viz. architectural challenges to implement cloud computing, rearchitecting applications to utilize massive parallel processing) between the chatter and the promises of cloud computing (elastic/scalable infrastructure).
• Look at cloud computing as another service delivery channel (in addition to in-house services and outsourced services)
• Advise to IT
  • Optimize your service delivery channels and impact to business
  • Map business attributes to the channels and manage a multi-channel environment efficiently
  • Plan for adaptive/flexible/scalable IT
  • Participate in the evolution of the cloud

A Q & A session with Sandeep Johri followed the keynote. HP is looking at applying cloud computng to consolidate data centers operated by EDS. There was a quick discussion around progressive CIOs who are intrigued by the cloud computing and are looking at applying the fundamentals “to radically deliver services at a vastly reduced price point”. <KS> I liked the characterization! </KS>

Selling the Cloud to Wall Street and Main Street

• This was an interesting session on cloud economics, especially from a technologist’s point of view !
• While SaaS is being adopted, the companies are spending 50-100% in sales and marketing. A very unprofitable business model with too much $ in the sales/marketng column
• The innovation in cloud domain should be “not selling software-software is bought!” Focus on bying cycles than selling cycles ! There was a discussion on managing leads – as an example one catalog company, on the web, receives 20,000 – 30,000 leads  and there is no way to manage them by hand. There needs a organic way to manage them!
• Another innovation is the business volume – sell many $25K than $1 million sales. Also subscription based models than one time sales
• Cost of customer acquisition is a very important metric
• In this regard (ales,marketing) cloud is bending the model than breaking the model
• Enterprises should learn the art of customer acquisition from the consumer side, the on-line strategy
• The cost of innovation has definitely come down dramatically
• Big companies will migrate to big/established clouds
• Which platforms to migrate to depends on the CG of the application – for example SFA would go to salesforce.com
• One challenge is the inter cloud connectivity allowing connectivity across best-of-breed platforms

The Marriage of Security and Cloud Computing

The keynote by John Maddison [Vice President, Core Technology Solutions, Trend Micro] was interesting.

They see cloud computing as a necessary platform for their applications. He was emphatic that without cloud they will not be able to manage the security threats we face in a few years. They use hadoop for processing now and plan to architect collection, processing and distribution of threat vectors around cloud infrastructure. They predict 233 million threat vectors by 2015 ! If we follow the current paradigm of downloading the signatures, our lines would be choked with ust that function!

<KS>It is comforting to know that clouds are the only way for this domain, on the other hand depressing to hear that we need that massive scale to keep up with viruses, bot-nets et al. May be we should drown our sorrows with MR during the happy hour !</KS>

Demystifying the Cloud

The talk by Dr. Vishal Sikka of SAP was another interesting keynote. Earlier, dusring one of the panels, someone mentioned that Oracle and SAP do not get loud; but Oracle has at least a good management … Ouch ;o( Vishal was a little mystified, but went on to give a good speech.

• We have achieved power, infrastructure and operation optimization
• But are far awa from integration, integrity and elasticity ! <- all cloud attributes
• Optimization is not a single dimension, different applications need different types of optimization (Slides 2 and 3)
• Slide 4 is a good illustration of the breadth of optimization required
• Slide 5 is a GMail page showing the breadth of elasticity. Different parts of the web page has different characteristic fed by appropriate infrastructures
  • The search area by the GFS, Chubby, mapReduce et al
  • The left side bar by a mail infrastructure and
  • The right advertisement bar by appropriate inference & data infrastructures

He completed the talk with an excellent slide on Timeless Software “Delivering this over containers that span multiple generations of technologies; Minimizing the cost and maximize the ease of its construction, deployment, and life-cycle management; In a landscape that is permanently heterogeneous”. <KS> Just beautiful, brings tears to the eyes of a technologist and well said !Their motto is “every activit, every user, in every business” !</KS>
In the following Q & A session with MR, he talked about “business by design” running “over a collection of processes over a clod infrastructure”

How would clouds evolve ? Mega clouds, cloud outsourcing and entrepreneurs using clouds

Ke customer opportunities ? Early stages of adoption of edge apps, major customers running mission critical systems on dedicated clouds. Need to manage cloud relationships very well. Too early to tell

Impact of cloud on data ? Issue is to maintain visibility/control. It is not that data is out of the enterprise, but need visibility. For example certain legal documents need to be destroyed at certain times, and enterprises need evidence that the content is in fact destroyed. <KS> I thought this is an interesting view. I also am of the view that a lot of enterprise data resides outside an organization – as part of backup, disaster recover plans, with vendors et al. </KS>

Understanding Enterprise Requirements for the Cloud

An interesting panel – three views

• CIO, government – Government regulations dictate data privacy, locality et al. A cloud operator cannot replicate data to other parts of the world – the CIO will go to jail ! But looking to leverage cloud for hosting public data as well as using public data (for example GIS) for applications. Clouds do give flexibility as she need to go to the legislature for more headcount and resources
• CIO, University – Most of the apps in cloud ! e-mail, CRM, alumni management, …. Feel that data security is more with clouds than can be achieved by a 40 person IT department. Clouds take low value complexity and they focus on the rest. Doing their own integration; so their infrastructure is the hub and the various clouds are the spokes. See increased liability due to higher vendor risk and vendor dependence on the operational integrity of the business. Per peeve on SLAs, termination clauses et al
• Director, web publication – Naturally web apps, mixture of clouds. Use clouds for various apps like surveys et al. Will do more clouds in the future

Finding Opportunities in The Cloud

A discussion session between MR and Navin Chaddha [Managing Director, Mayfield Fund]

• Cloud is evolutionary
• Makes sense for Amazon to get into cloud space because the need the infrastructure anyway
• Start-ups should not focus on capital intensive infrastructures, but have opportunities on cloud services like management, billing et al on mega clouds like Amazon’s. ie focus on building value (tools,applications) over basic cloud infrastructures
• Clouds – definitely provide capital efficiency for start-ups and consumer web 2.0 apps
• One interesting question was on data ownership and legalese – Would you rather have your lawyers answer a subpoena than your cloud providers lawyers ?
• Clouds – consumerization of It not commoditization of IT <KS> Interesting and very accurate. I am a big fan of clarity, in terms of concepts, because they have long implications on architecture and business models</KS>
• Large enterprises will go to mega clouds. Cloud will have bifurcation – large and small
• Currently there is no major catalyst for cloud computing

In Short

Good conference in the business sense. Excellent keynotes, learned a lot. I felt that the breakout sessions were a little anemic and the breakout presentations we more marketing – YMMV

Clouds ready to chuck training wheels ? A view from Users, Vendors and VCs

Exec Summary:

I attended the “Cloud Computing and Beyond: The Web Grows up (finally)” conference hosted by SDForum in Santa Clara. Well done and was very informative. I was looking for the “beyond” part – pointers like “what are the barriers for wide adaption of cloud computing (especially in the enterprise)?” and “What are the opportunities?”. And got a lots of good ideas (as you will see in this – slightly long – blog!)

As a quick summary, I was impressed by the crispiness and clarity of what the users want from the current crop of cloud builders, especially Amazon. The shortcomings include latency variability, no stated SLA, proprietary interfaces and opacity. There are some requirements for seamless hybrid clouds, with spatial data and process flexibility between both sides of an enterprise firewall, ability to download control plane software (a.k.a the cloud OS) into on-perm servers (basically turn a set of arbitrary servers on this side of the firewall into a cloud, as needed!) as well as the declarative-policy based administrative control to achieve the “fluidity”. Projecting from this microcosm of users, I think, a few changes are in order for the cloud builders, like Amazon.

In short, the keynotes were inspiring, the vendors a little controversial, the users very passionate and somewhat frustrated and the vcs – well they acted like vcs – vague yet confident !

The Keynotes:

The first keynote, by James Staten of Forrester Research, set the tone for the day. He did a good job in framing the domain, even has a good definition!
The points I was able to grasp:

• Cloud Washing – a good term indicating the addition of cloud to an product in the market, especially from old web hosting companies.
• One implication of the cloud computing is the changes in business interactions including create markets faster (and the flip side fail faster), ability to try out newer concepts faster e al
• Business wants a place to experiment with fast integration, looser IT restrictions and faster responsiveness while It wants predictability, stability and early notice. So there is a level of impedance mismatch and cloud computing fill that void very well!
• I liked Forrester’s definition – “Cloud is a pool of scalable, abstracted infrastructure that hosts end-use applications, billed by consumption”
• Value proposition of clouds “Two persons, a laptop and a credit card = Web presence !”
• There are four major customers
  • Start-ups/Individual accounts
  • Gaming & Entertainment Companies
    • This was news to me ! It seems maintaining a game infrastructure takes less resource but spikes when they introduce a new game. Excellent use case for clouds
    • Looks like even Blu-ray movie releases have this challenge – Iron Man DVD release brought down Paramount’s servers completely (here and here) because the discs download content as soon as they are inserted into the player, which caused a stampede. Many folks returned the discs thinking they are defective ! ep an Amazon cloud could have helped here !
    • Another use case is promotion sites for new movies (again very temporal!)
  • Small businesses (web presence)
  • Enterprises (very small percentage)
• Challenges to meet enterprise needs
  • The current cloud offerings do not meet IT needs
  • Not easy to control by IT
  • Don’t support general as well as IT specific practices

Lew Tucker from Sun gave a keynote and had good points:

• Salesforce was able to turn an application into a platform
• Extending Carr’s book, Lew is of the opinion that distribution was the key for the adoption of electricity; a lesson the cloud community should consider

Another keynote was from Dr.Jayashree of IBM. Very insightful and interesting presentation. Some highlights:

• IBM’s cloud centers are interesting – they are actually listening posts ! Customers can come in, elaborate their business problems, prototype and then see if cloud fits their needs – a very neat way for IBM to engage and work with their customers, especially in a new domain !
• Use cases IBM is seeing from customers include innovation enablement, software development facilitation, virtual clustering and web 2.0 data intensive processing

I had to leave during the keynote by Russ (HP)

The Users:

Good discussions on what is lacking in the current offerings – especially by amazon.

• Users still do not trust their data in the cloud. Some form of assurance – certification, best practices as well as audit might help. As I had mentioned in one of the e-mails, the two areas viz:
  1. Security and reliability – I think we need some form of certification based on (one or all of) best practices, audits, standards and specifications.
  2. Compliance – this is a legal issue as well. Again, need some form of industry audit/certification that makes a cloud infrastructure as “legally secure” as an internal infrastructure
• One surprise mentioned was how much work need to be done with Amazon’s offerings
• Another was the reliability issue – VMs go down and corrupt databases
• The flip side is that the limitations posed by Amazon’s offering has forced towards a better architecture (based on constraints like limited size VMs that can disappear) that scales well and keeps with organic growth. Otherwise designs would have migrated to bigger machines leading to not-so-scalable solution.
• Latency seems to bother a few folks, especially the variability. For many the VM movement seems arbitrary
  • Looks like there are opportunities for optimization in VM distribution & migration that optimizes variability in latency – for example all related VMs nearby – based on interactions optimization. I think this is also related to affinity mentioned in a couple of papers.
  • There was a discussion around the use of memcache and associated latency. I quite didn’t follow, but have a feeling this might be an area Amazon should look into
• Many, in the users panel, mentioned that Amazon is very opaque and that adds to the unpredictability nature of the latency and response. Of course, for a successful abstracted cloud, one should not know of the underlying implementation at all; as that will change. Moreover Amazon is constantly improving the control plane and the protocols and algorithmics running the substrate (I hope so)
• Users are at a stage now where they expect clear statement of SLAs and contractually stated variability. Amazon would be wise to heed this insight from the users, especially as many are about to negotiate on this point! Especially as the users have confidence in Amazon “Jeff’s team will be a lot better at running a cloud than we will ever be”

The Vendors:

The vendor panel was very interesting. Again, good insights. Some notes form the vendor panel discussion:

• Cloud infrastructure should be viewed similar to the communication infrastructure (a view I also share)
• Cloud control plane software is in a different end of the spectrum
• One insight from the vendors “Build your app for the channel”
• Clouds provide an opportunity to innovate at the application layer
• Architecture portability rather than application portability across different offerings
• Couple of provocative comments came from the vendor panels (Jason Hoffman of Joyent)
  • We need the ability to seamlessly transition between off-perm and on-perm clouds. For example one should be able to do dev and test using Amazon’s clouds and when the time comes for the production environment, one should be able to declaratively specify what components should reside where; and the same goes for data as well! Amazon then downloads it’s control plane into the on-perm servers as needed and manages this distribution seamlessly! Moreover one should be able to change the mix depending on the nature of the application, the compliance and regulatory climate one is in as well as other business policies.
  • I think the “dev-stage-prod” cycle of traditional architecture is replaced by “dev-cloud tryout-cloud-stage-prod” cycle with more dynamism as to the on-perm-cloud/off-perm-cloud distribution of apps.
  • There is also the mixing of namespaces, IP and VLAN mobility, load balancing across VMs which are in very different LANs and a host of other contextual complexities to be tackled in the cloud control plane – all of which we had made very strict assumptions in the traditional architecture! I repeat the axiom “robust chaos rather than brittle determinism” ! No more Newtonian machines !

The VCs:

Had some good discussions.

• They agreed that cost saving is not the essence of cloud computing (I also share this viewpoint. Clouds are about newer business models, business capabilities and so forth)
• There was a good discussion on how CDN was commoditized and whether clouds will follow the suit
• The see clouds as virtualizing the application infrastructure
• They are looking for companies that address long term pain points as well as have a short term promises

In Conclusion:

All in all, a good conference. I got what I was looking for. Couple of quick notes:

1. The presentation from GoGird looked, out of place (sorry guys, I like you, but you need to be relevant at that context) – a very anemic sales pitch, no clear value proposition (their claim is that they run windows and are cheaper than AMZ – both very transitionary (AMZ will run Windows soon and pricing is relative anyway) and looked like a “cloud washing” by a hosting company looking for relevance as a cloud provider. I could be wrong(I hope I am!)
2. The enterprise panel really didn’t address the “Challenges and Opportunities”
3. Lots of discussion around disaster recovery as a value proposition for clouds. Don’t know how the discussions degenerated to this; I thought was a doozy ;o) Not that disaster recovery is a good segway (once folks get used to data outside their vision, they will be more comfortable with clouds in general) but disaster recover as a cloud feature is way far off, IMHO.
4. I have a nagging feeling that VCs really do not get clouds – except as a “hot” area. While I have to assume that they know much more than I do and have a far better judgment, I didn’t hear any insightful comments. May be they are intentionally vague.
5. Joyent’s Jason had some insightful comments;he also had a lot of controversial ones. It is interesting that the ones I agreed with him – I had complete agreement; and the rest I totally disagree; no shades of Grey !
6. I have more pages to transcribe ;o) Will update this blog with a few more details …