Friday, March 21, 2008

The need for standardization!

I was sitting in the airport lounge in Frankfurt when I came across an article in Time magazine. It was a story about a British real estate development company that built a building for a group of artists “fashioned from 20 recycled containers in London’s Docklands district” back in 2001 and had since constructed another 27 buildings from these old containers. In the London suburb of Uxbridge, “a 120 room Travelodge hotel is being built out of 87 (of these) steel crates, stacked and snapped together Lego-like into two nine-story towers.” The picture I have included here is of the construction of this building where the installation of the containers is clearly visible.

I wasn’t working in IT back at the birth of containerization, in fact I wasn’t even at school yet, when in 1955 the “world’s first truly inter-modal container system used the purpose-built container ship, the Clifford J. Rodgers” sailed between Vancouver, British Columbia and Skagway, Alaska. It carried 600 containers that had been delivered dockside on trucks, that were then “unloaded onto purpose-built railway cars for transport north to the Yukon.”

But I did join Overseas Containers (Australia) Limited (OCAL), a subsidiary of P&O, in 1972, just 3 years after the first container ship had arrived in Sydney. The IT department had to make sure tapes of the ships manifests could be processed and be made available for customs agents. I learnt a lot about bill-of-ladings, tracking empty containers, full container loads as well as less-than-full container loads (FCLs vs LCLs), etc. And I learnt about managing the security of the containers and ensuring that there had been no tampering along the way! The picture alongside here is of the Encounter Bay, the first of the new class of "Encounter Bay" container ships to arrive in Australi in 1969.

The time I spent working in the container shipping business really reinforced for me the value that comes from standardization. The industry would have been a total disaster if different trade organizations had held out for their own container sizes – or if we had elected to support both an imperial and metric container. The duplication, perhaps triplication, of port facilities alone would have driven up the prices so much that shipping in bulk and general cargo ships would have reappeared as an economically viable alternative.

When it comes to standardization, we are beginning to see the same revolutionary approach to packaging appear in the computing industry. It is already well on its way when it comes to storage, and now it’s all about blades! From the blade chassis, to the blades themselves, it’s very much become a drive to standardization. And the reason for much of this focus is to drive down costs, while increasing the options we have.

Looking at them separately, the lifecycles of the chassis and the blade will be very different. Decisions taken with the design of the chassis, and the technology it supports in terms of connectivity and the storage that can be hooked-up – what can plug into what – will influence the generations of blades that follow. Chip designs continue to evolve rapidly and that’s no surprise for any of us. But what really many of us want to see is the ability to integrate new generations of these chips into a single chassis so that the price benefits can be immediately realized. Over time, chassis changes will be introduced into the marketplace very cautiously, and I fully expect to see vendors support combinations, or hybrids, so that transitions to new chassis can be easily undertaken with minimal to no downtime.

However, a lot of work is still required with the blades themselves and with key infrastructure components, before they are a truly standardized package. For instance, the industry really does need to move more aggressively and produce a single interconnect technology for clustering. As much as I am a huge fan of ServerNet I cringe at the sight of specialized mezzanine, or daughter, cards being added to today’s blades. If we are to see a standard blade become available, that supports NonStop, HP-UX, Linux, Windows, etc. then just one interconnect technology needs to be integrated onto the blade and be available for any OS to exploit.

The recent decision by IBM to use InfiniBand (IB), on their new z10, as the interconnect technology for accessing all types of controllers, has to be a big first step in the right direction. The “footprint of ServerNet is all over IB” I have written previously, quoting a senior HP engineering manager, but it’s true and we need to be reminded of it. I believe that future iterations of IB need to be integrated into every blade so that an industry-standard, universal, blade appears. NonStop can be easily enhanced to take advantage of it and for Unix, Linux, and Windows developers looking to support cluster configurations, it’s an ideal technology.

The costs for each blade would come down significantly, and the need to constrain the running of an OS to a specific blade instance would disappear. The flexibility that this would give IT would be enormous as they could simply plan on adding “x” number of blades to their collection of blade chassis each year, without the need to spend too much time looking at which blades need to be allocated to which OS. Solutions that best met the needs of a company can now be deployed without any real need to worry about the OS required. Nonstop could move a lot closer to center stage, and its attributes better utilized, as its need for unique hardware subsides.

But perhaps an even bigger challenge lies in standardizing the infrastructure middleware required to support all of this. Data bases don’t provide such flexibility today out of the box. For instance, a data base running on Windows cannot be accessed by an application on Linux as if it were MySQL. Even though SQL is an industry-wide standard, it doesn’t provide any interoperability flexibility. It’s almost as if there’s a need for a parallel backplane technology – not just the physical backplane in the blade chassis, but a logical or virtual backplane that transparently interconnects the data bases. And this is not data virtualization either!

For some time now, GoldenGate has supported data base tiering outside the box. Sabre did a lot of pioneering in this field and today, has a powerful configuration in place across a mix of NonStop SQL and MySQL. Outside of the travel industry, I have seen other industries begin to look at this model – from stock exchanges to manufacturers. But bringing this technology inside the box needs to become a serious consideration – if for no other reason than the breakthroughs that will come from standard blades and blade chassis may be offset by additional data base complexities that materially impact the overall value proposition that should arise from any standardization of blades.

Once containerization began to take hold, it wasn’t long before its versatility attracted non-traditional shippers. Shipping fresh produce across the globe had always been problematic with many shipments arriving at their destination spoiled. However, new container designs appeared that would take a “clip-on” refrigeration unit that, when lifted onboard the new generation of refrigerated container ships, or “reefers”, plugged straight in to the ships central refrigeration system. Containers would simply have their clip-on’s removed as they were loaded onto the ship, and then reapplied as they were unloaded.

Perhaps we need clip-on’s for data base? Perhaps the solution is to clip-on a software module that keeps the newly deployed data bases completely synchronized. It is not that big of a stretch to consider such a development as being close at hand, as most of the underlying technology already exists. The premier products in the change data capture field could play a major role in providing these essential clip-on’s.

The arrival of standards for blades and blade chassis, and perhaps even with data base access, will continue to excite me. I see tremendous potential to reduce costs while opening up the options and broadening the range of choices available. Earlier this year I wrote about my three wishes but, in reality, there has to be changes at this level before any of my wishes can materialize. But I suspect, as we watch the technology announcements unfold this year, we will begin to see much of what I wish for take on real substance and a world of choice open up. And who can tell, perhaps the demand for NonStop may move it closer to center stage faster than any of us expected.

And what were those three wishes, you may be ask? Take a look at February 12, ’08 “‘My Wish’ for NS Blades!”

1 comment:

Marty Turner said...

Yup, well having worked with InfiniBand in its early days (when it came to life as a merging of Future I/O and NGIO) I only have respect for it as a switched fabric architecture (see for specs etc.). And yes it's got ServerNet all over it because one Dave Garcia (then working for Compaq labs I believe) as on one of the InfiniBand working groups. One can certainly call IBand "commoditized ServerNet" for all intents and purposes. BTW I always thought that NonStop missed the boat by not making SNet more public (again, even though parts of it crept into IBand).