If I interpret the comments on my last post correctly, both online and offline, a small number of you felt that I’d been unduly strong in my bias against the “private cloud”; it sounded like you thought I’d been drinking too much of the Kool-Aid since joining Salesforce.com a year ago this weekend.
Actually, my bias against the private cloud is around a decade old. And it stems from experiences I had during my six-plus years as CIO of Dresdner Kleinwort.
First and foremost, I think of the cloud as consisting of three types of innovation: technology, “business model” and culture. Far too often, I get the sense that people concentrate on the technology innovation and miss out on the remarkable value offered by the other two types of innovation. In this particular post I want to concentrate on the business model innovation aspect.
Shared-service models have been around for some time now, they’re not new per se. At Dresdner Kleinwort, we implemented shared-service models wherever relevant, sometimes within a business unit, sometimes across business units within a business line, and sometimes across the whole company. The principle was simple: investment and operating costs (the “capex” and “opex”) for the shared service would be distributed across all the consumers of the shared service according to some agreed allocation key. Sometimes it was a simple key, like headcount. Sometimes it was predefined each year at a central level, as was the practice with “budget” foreign exchange rates. Sometimes it was hand-crafted by service, involving long hours of painful negotiation. Sometimes it wasn’t even agreed, just mandated from above. One way or the other, there was an allocation key for the shared service.
Dresdner Kleinwort was part of Dresdner Bank, and Dresdner Bank was wholly owned by Allianz. There were shared services at the Dresdner level, and at the Allianz level. So there was a whole juggernaut of allocations going on, at multiple levels.
And God was in His Heaven, All was Well With the World.
Until someone wanted to leave the sharing arrangement.
At which point all hell broke loose.
Because the capex had been spent, and the depreciation tail had to be allocated to someone. If the number of someones grew smaller, the amount allocated grew larger. This wasn’t just about capex; not all of opex was adequately volume-sensitive, so similar effects could be observed.
“Private” models of shared services were fundamentally zero-sum games: the institution coughed up all the capex and opex, and the institution had to allocate all of it. Regardless of the number of participants. Sometimes there was scope for some obfuscation: there was a central pot for “restructuring”, and all the shared-service units ran like hell to reserve as much of it as possible every time the window opened for such a central pot. If you were lucky, you could dump the trailing costs left by the exiting business into the restructuring pool, thereby avoiding the screams of the survivor units. But it was an artificial relief: the truth was that the company bore all the costs.
A zero-sum game.
Shared resources have costs that have to be shared as well. If the only people you can share it with are the people in the company, then the zero-sum is unavoidable. Things are made more complicated by using terms like capex and opex, by choosing to “capitalise” some expenditures and not others, by having complex rules for such capitalisation. Such worlds were designed for steady-state, not for change.
We’re in a business environment where change is a constant, and where the pace of change is accelerating. So there’s always something changing in the systems landscape. Business units come and go; products and services offered come and go; locations and even lines of business come and go; and entire businesses also come and go within the larger holding company structure.
Change is a constant.
So with the change comes even more pain. Lists of “capitalised” assets have to be checked and cross-checked regularly, to validate that the assets are still in use; at Dresdner these were called impairment reviews. If not, the remaining depreciation tail of the “impaired” asset has to be absorbed in the next accounting period.
What joy. [Yes, dear reader, the life of a CIO is deeply intertwined with the life of a spreadsheet jock].
In many respects, the technology innovation inherent in the cloud was foreseeable and predictable. Compute, storage and bandwidth were all going down paths of standardisation, to a point where abstract mathematical models could be used to describe them. As the level of standardisation and abstractability increased, the resources became more fungible. That fungibility could be exploited to change the way systems were architected, higher cohesion, looser coupling, better and more dynamic allocation and orchestration of the resources.
The business innovation in the cloud was, similarly, also foreseeable and predictable. The disaggregation and reaggregation made possible by the standardisation and virtualisation would allow for different opportunities for investment and for risk transfer.
Now it was no longer a zero-sum game. The company that spent the capex and opex took the risk that there would be entrants and exits, high volumes and low; the technology innovations were used to balance loads and fine-tune performance; the multitenant approach often led to lower licence costs, and these could be exploited to defray some of the continuing investments needed in the balancing/tuning technologies.
Individual business units and lines and even entire companies no longer had to carry out impairment reviews for such assets. Because they didn’t “own” the assets: the heart of the cultural innovation was the change in attitudes to ownership.
The private cloud proponents have sought to blur the lines by bringing in arguments to do with data residency.
Data will reside where it most makes sense. Sometimes there are regulatory reasons to hold the data in a particular jurisdiction. Sometimes there are latency reasons to hold data within a particular distance limit. Sometimes there are cultural reservations that take time to overcome. The rest of the time, data can be held wherever it makes economic sense.
Serious cloud computing companies have known this, have been working on it, and will continue to work on it. The market sets the standard.
Code, on the other hand, particularly multitenant code, has no such residency requirement. Unless you happen to ask someone whose business model is to charge licences connected to on-premise processors.
Change is a constant in business life. The cloud is about change. The business model of the public cloud is designed to make that change possible, without the palaver of impairment reviews and capex writeoffs and renegotiation of allocation keys and and and
Which is why, in principle, the private cloud makes no sense to me.