Monday, April 11, 2011

A question of standards.

I'm all in favour of standards for the "cloud" world because such standards should enable and accelerate innovation in IT through componentisation effects as well as encourage the formation of competitive markets of compute utilities

However, that said, I'm against standards committees and the concept that open standards (as in APIs & data formats) are enough to create the portability required for competitive markets.

In the former case, the market will decide the standard and the job of any standards body should be to rubber stamp what is an existing practice. Unfortunately, standards committees are often used as vehicles to promote specific vendor interests and in many cases their efforts are counter productive. For example, whilst IPX/SPX was the committee approved standard, it was TCP/IP which won the marketplace battle. The only impact that approving IPX/SPX as a standard had was to temporarily slow adoption of TCP/IP in some quarters and hence inhibit innovation.

Standards should have a positive effect but defacto has to precede dejeure. There are some exceptions to this but they are exceptions.

In the case of open standards, a competitive marketplace requires multiple providers, access to code and data (ideally with syntactic interoperability) and semantic interoperability of services. Whilst open standards provide part of the solution, it is critical for reasons of semantic interoperability that a common reference model (i.e. running code) is provided.

Since, we're talking about a world of service (and not feature) differentiation for activities which are fundamentally commodity by nature and hence suitable for utility service provision, then the obvious solution is an open source reference model as the standard. Potential examples of such would be the OpenStack effort.

Such open source reference models would ideally exploit the dual nature of GPLv3 which is both restrictive in the product world but simultaneously permissive in the service world to create a functioning marketplace. Unfortunately, many vendors promote open standards (as in APIs etc) as the solution to these problems of portability and hence describe their systems as open when they're quite clearly not.

So, in general :-

  • Are standards good for the cloud?
    Absolutely, it'll encourage innovation through componentisation effects.

  • Are standard committees good for the cloud?
    Generally no. At best they should rubber stamp market chosen approaches however in reality they're more likely to get in the way or slow progress by promoting vested interests.

  • Is portability between providers important for the Cloud?
    Absolutely, it's the route to formation of competitive marketplaces and reducing outsourcing risks.

  • Will open standards provide the portability needed?
    Only in the most trivial cases but not for the vast majority of activities. Open standards are necessary but they are not sufficient to provide the portability required. The idea that open standards alone will achieve this will inhibit the formation of competitive markets.

  • Is open source essential for the cloud?
    Absolutely, the formation of competitive markets without loss of strategic control to a specific vendor depends upon the provision of open source reference models as the standard.


Geoff said...

You need to get your facts right. IPX/SPX was never a "committee approved standard": it was a commercial product from Novell. TCP/IP was the result of many "committees": various groups under the auspices of the IETF, all committed to "consensus and running code".

There are plenty of examples of committee-developed failures: OSI, DCE, WS-splat, even XML. But IPX/SPX wasn't one of them.

Geoff said...

Egregious errors aside, I agree with your thesis, of course!

swardley said...

Hi Geoff,

Good spot. I was actually referring to the protocol wars and how ISO attempted (and for a time succeeded) to enforce the OSI model on government networks and in effect ban TCP/IP.

I know why I mixed my stories here, whilst there were OSI compliant technologies, if I remember correctly IPX/SPX was often touted as being more OSI compliant. In fact, the general rule of thumb seemed to be anything but TCP/IP. Even in the final death throws, OSI was throwing up models such as "minimal OSI" just to try and stay in the game etc.

So yes, fair point, mixing my stories terribly. I should have referred to the whole ISO vs IETF example and how eventually the market won and how ISO slowed adoption for a time in some industries.