2009年6月2日火曜日

A Vision for Oracle’s Cloud Platform: The Red Cloud

OracleのCloud Computing戦略についての考察。 

Tony Nemelka氏はPeopleSoft、Epiphany、Adobe等の日本市場責任者として日本市場に非常に詳しく、Oracle JapanがForce.com同等のサービス事業を必要としている事が指摘している。  日本のSI事業が特にこの市場を狙っており、特にソフトウェアベンダのSaaS事業への転換を支えるPaaS事業者として成長する可能性が高い。 

Oracleは自社のPaaS事業に向けて必要なツールは揃っている、と指摘している。  実際にサービスとして展開するまでは相当の開発や統合が必要なのは否めないが、下記の買収がキーになる、と述べている。
- ハードウェア:  Sun Sparc
- OS:  Sun Solaris
- 仮想化技術:  Sun
- DB:  Oracle、MySQL
- APPサーバ:  BEA
- BI:  Hyperion
さらに、これらのツールを通して、恐らく下記のような事業を順次展開するのでは、と予測している
- Storage:  Sunの持っているZFSを通したストレージ事業
- Identity Management: Oracle、Sun両社のこの領域の強みが生きてくる
- Business Intellegence: BEAの技術
- Integration:ハード/OSを組み合わせたシステム事業
強いてあげれば、Multi-Tenancy技術がまだ抜けている。  これについては自社のDB技術を通してデータベースの仮想化、マルチテナント化を提供する戦略をとるのでは、と予測している。 

--------------------------------------------------------------------------------

A Vision for Oracle's Cloud Platform: The Red Cloud


Helpstream's CEO (my boss), Tony Nemelka, penned a great piece on the Helpstream Blog about what the Oracle/Sun acquisition might mean.   A lot of attention has been focused on the potential for negative outcomes.  Will Oracle kill MySQL or at least damage it with worse than faint praise (fascinating post by fellow Enterprise Irregular Josh Greenbaum)?  Other note (quite rightly) that Oracle can't really kill MySQL because of its Open Source basis.  I found WordPress founder Matt Mullenweg's post to be particularly eloquent of this.

What I found intriguing about Tony's post was the more positive scenario it envisions.  Read the post, but let me summarize for purposes of this discussion.

Tony was recently in Japan and has a long history there having been an executive for PeopleSoft, Epiphany, and then Adobe in charge of the region.  Needless to say, his contacts are pretty high up the food chain, so they know what's going on.  In that world, the System Integrators are the gatekeepers for the market.  They're very powerful, and the interesting discovery Tony made is that they absolutely love Force.com.  It's not hard to see why.  The SaaS model squeezes the SI ecosystem.  The normal meat and potatoes business around just getting on-premises software installed is greatly reduced.  The business of just keeping the lights on is almost non-existant for SaaS.  Yet SI's have a lot to bring to the table.  A good SI often understands the Domain, its Best Practices, and the key Business Processes better even than the software vendor.  Having access to a SaaS platform makes it possible for the SI to turn that valuable knowledge into product which can then be sold.  That's why having a platform on which to do that is so important to them.

Tony goes on to speculate that Oracle is picking up the components necessary to create such a platform.  If nothing else, Oracle's Japanese SI's are screaming that they need one.  I have to imagine SI's everywhere are grokking the essential value of a platform to the SaaS ecosystem.  There's nothing about the Japanese market that would make that a unique requirement there.  So far, Oracle is really stuck in that department.  I suppose they would argue the coming Fusion represents such a platform.  At the same time Sun, like every big hardware vendor, was hard at work crafting a Cloud strategy.  They all know the Cloud is a tidal wave that can profoundly impact their businesses.  The Cloud represents the federation of many smaller deals into fewer gargantuan deals.  One over simplified way to view it is as a whole new sales and marketing channel.  Failing to suit up for the game guarantees a loss and the stakes are high.

Some I've talked to say that Oracle just "doesn't get it."  They don't believe in SaaS, they don't understand SaaS, and they can never execute this kind of nuanced strategy until it is way too late.

The idea that Oracle wouldn't try or wouldn't deliver anything is not something I'd want to put too much money on.  It'll probably take them longer than anyone would like, or they might surprise us too.  Even Oracle can't really fight the Cloud/SaaS tidal wave.  Remember that the guy at the top believes it in, don't forget that Larry Ellison has put his money into SaaS companies many times.  It happens to be inconvenient at the moment for the various financial metrics Wall Street cares about for Oracle to switch wholesale to SaaS, but even that is something they can manage over time.  Also, we know Sun was working hard on their own Cloud strategy.  Suddenly there is a lot more Cloud DNA coming into Oracle.

I suspect Oracle's vision of what a PaaS (Cloud Platform) is will be a lot different than what you or I might choose.   If nothing else, it won't be a clean sheet of paper approach.  Let's think about what it might be.

First, I would expect the initial version will not be multitenant.  Multitenancy as it is delivered today is too deep in the guts of the application.  It forces too much change to architectures, which creates too much adoption friction for a platform at the outset.  Oracle will want to deliver the promise of running any app on their Cloud Platform, or at least any app built to run on their now very comprehensive stack.  It spans hardware (SPARC et al) to OS (Solaris) to DB (Oracle/MySQL) to App Server (BEA) to Business Intelligence (Hyperion) to <you get the idea, phew!>.

Second, if not multitenant, what?  Think virtualization.   Don't they need to buy VMWare to get that?  They may buy VMWare, but they don't need it for virtualization.  Sun Solaris has a wonderful virtualization capability built right in.  I've used it before to create a SaaS application and it works extremely well.   Imagine an Amazon-like capability to start up these virtual Solaris machines.  If Oracle is smart, and they usually are, you'll be able to start up virtual "appliances" in the Red Cloud (my new name for Oracle's Cloud) that deliver database, app server, and many other functions.  Consider:

-  Storage:  This one is obvious.  Sun has a big storage business, Jonathan Schwartz has blogged about their great ZFS storage technology, and so Oracle can easily deliver the "Amazon S3″ storage piece of the Red Cloud.

-  Identity Management:  Control over who logs on to the Cloud.  Sun is strong in Identity Management, for example.  Oracle already had a business there.  The combination of the two would create a leader to rival IBM and possibly be the world leader.  Having Identity Management built into the Red Cloud would be a decided work savings over Amazon, for example.  And Identity Management is one of those things on-prem apps are used to farming out to another module.  Hence it would facilitate their migration to the Red Cloud.

-  Business Intelligence:  This is another of those modules everyone wants to OEM instead of having to build for themselves.  It's an ecosystem component, like Identity Management, ETL, and a host of other things.  Oracle can again deliver a virtual appliance in the Red Cloud that makes it simple to connect.

-  Integration:  This will be essential to making the Red Cloud Appliances work.  But when you have one vendor that controls so much of the stack, they should be able to make it work better than anyone else.  This is where the vision of being the "Apple for the Enterprise" can best be seen.  This is where a lot of the Fusion work, as well as work from BEA could tie in.

OK, that's a pretty darned impressive first tier vision if you ask me.  It's taking the old pitch about buying a suite from one vendor and ratcheting it up significantly in terms of scope.  Not everyone will buy into it, but Big IT might just need something like this before they can start moving to the Cloud.   If Oracle can deliver such a thing, it will be an enormous business.  Someone I was talking to recently said they thought Microsoft was Oracle's ultimate acquisition target, but I think Larry has a shot at rivaling much bigger entities if he can execute.  The IBM's and HP's are starting to appear on his horizon.

But there are some warts.  It will be a hodgepodge unless some Fusion-like glue can make it all fit.  It will be an enormously complex offering to market and sell.  It will be nearly impossible to take in the scope of it or understand all of it.  All of the Enterprise complexity that Big IT loves, but that SaaS has tried to ameliorate will be there.  Granted, this is better than the old on-prem complexity.  Oracle can deliver at least some of the SaaS advantages.  But what about cost?  Can we really get the SaaS cost advantages without multitenanacy?  In a word, "No!"  I just wrote an article about the pitfalls of thinking virtualization is a substitute for true multitenancy.  I stand by every word I said.  But Oracle has a unique opportunity to virtualize multitenancy itself, not in the first iteration of the Red Cloud, but in later iterations.

Whoa!  What the heck is Bob on about now?  Virtualizing multitenancy, what does he mean?

What I mean is a mechanism whereby single tenant applications can be made to have all the benefits of multitenancy without radical architectural change.  There are two ways this can happen.

The approach is to have the database itself virtualize multitenancy.  The most popular model for multitenancy is what I call columnar:  a column in the database tells which tenant each record belongs to.  A suitable feature set in the database server can completely automate this and make it radically simpler for the application to work along this model.  A second common model gives every tenant their own set of tables.  Here again, special support in the database can radically simplify the implementation.  Note that Oracle already does partitioning (they may have invented it, but I am not sure of that), which is a feature that makes a bunch of physical tables look like a single table and that greatly improves scalability.  So now Oracle could deliver a Multitenant DB Appliance in the Red Cloud.

The second approach attacks the costs of being single tenant.  In my post on the subject, I talked about the idea of fixed costs and variable costs.  The difficulty is that the database server uses up a bunch of system resources on each virtual machine even if there is no data loaded into it.  This is because the database server is unaware of the other database servers in their respective virtual machines.  They are isolated.  But what if they were aware of each other and could pool those fixed cost resources to reduce the overhead?  What if the operating system itself facilitated this?  Those fixed costs could be dramatically reduced.  Moreover, if a comprehensive integrated feature set was aimed at reducing the cost of administering such as system, we would likely start closing in on true multitenant efficiency levels.

That would be my vision for Oracle's Red Cloud.  It would be a first-class Cloud platform, although much more Enterprisey and Old School than the New Cloud Age SaaS and Amazon-style Cloud visions.  Oracle and BigIT will view that as an advantage.  The biggest challenge in all this will be execution.  It is a gargantuan task on a level Oracle has never delivered before.  It involves both coordinating a lot of existing parts and pieces as well as delivering some genuine innovation (Virtual Multitenancy will be non-trivial!).   They may not be able to pull it off.  But the stakes are very high, and they will have years to work at it.