Are you getting green-weary? It would not surprise me if you were because everywhere one turns these days there is an another admonishment of the IT industry’s un-greenness or some announcement by yet one more company pitching itself at green sainthood. And when that company is IBM a little cynicism can be allowed, and when the system involved is the mainframe then a certain level of hilarity can readily be anticipated.   But the green argument is one of the strong suits being put forward by IBM in its promotion of the recently launched z/10 mainframe system, and numbers being quoted by the company are sufficient to make one stop giggling and take a bit of notice. They are, in short, very impressive, particularly if they really stack up over the long haul of a system lifecycle. 

The targets being quoted for the z/10 include an upto 80% reduction in energy consumption and costs, an 85% saving in floor space, which should prove attractive anywhere the price of property rates highly as a budget item, and an 80% reduction in associated datacentre labour costs through reduced support requirements. Those labour costs, of course, not only include direct salaries but all the attendant costs, not least of which is the energy consumed by their need for lighting, air conditioning and the rest. 

OK, so the important question here is what all these reductions are set against. If it was the previous z/9 mainframe that would be impressive but irrelevant to all but committed mainframe shops. In fact it is set against a datacentre of Linux servers. A single z/10 can run up to 4,000 virtual Linux servers which, using any other technology, equals a large number of racks. What is more, they can be run within the z/Series operating environment, which requires fewer sysadmin staff anyway, typically no more than a handful. The headcount for an x86 server datacentre is typically estimated at one sysadmin for every 10 servers, so 4,000 would require some 400 staff on that basis. 

Of course, the real question is whether those 4,000 virtual Linux servers can function as well as real x86 boxes. Only time and the evidence of users will prove that one way or the other. IBM claims that they can operate the same generalised workloads, and have them run with the higher levels of separation and security that are inherent parts of z/Series partitioning. As the z/10 now sports a new, 4.4 GHz processor, which will also be used at the heart of the IFL (Integrated Facility for Linux) processor specifically designed to process Linux code, it will be interesting to see how well it compares running a single-threaded Linux application that would normally run on an x86 server dedicated to its needs. 

But even if it does poorly, it will need to be very poor, or such an application will need to be particularly important to the business, for it to lose the overall economic argument – namely that IBM estimates it can cut the total operating costs of 3-4,000-server datacentres, compared against a handful of z/series machines running z/VM and Linux, over a five-year lifecycle, by more than half.  

That is one hell of an argument, I want users and vendors to argue it, but it is one you have to sit up and notice.

A recent announcement from Intel had, at face value, absolutely no interest at all for anyone concerned with enterprise IT infrastructures – unless, of course, they were also totally committed PC gamers as well. The company was announcing the Dual Socket Extreme Desktop Platform, previously codenamed (for reasons probably best left unasked) `Skulltrail’. This is two quad-core processors mounted together in a single package, making an eight-core (oct-core?) processing unit. 

This, apparently, will be the bees knees for gaming on a PC, and is the latest, higher performing implementation of the dual quad-core approach the company first showed at last year’s Consumer Electronics Show. But for those interested in infrastructure, the announcement has some other connotations, for it should sound familiar. For any of them that use x86-based servers – and that will certainly mean most of them – there should be the memory of the announcement from Intel, this time last year, of Clovertown. This was a `quad-core’ processor made up of two dual-core devices in a single package, targeted that time at the server vendors. It was followed, later in the year, by the next step along the roadmap, a single chip quad-core device. 

One swallow may not make a summer, but two may indicate a strong hint of it and a repeat by Intel of this development tactic, putting two existing multicore processors together in a package, does suggest the blindingly obvious. This is that later this year a single chip eight-core x-86 device can be expected. That, in turn, will be followed by a dual eight-core and the subsequent 16-core processor, probably during the second half of 2009. 

Following the pattern onwards now becomes easy: the number of cores per chip will double every year, which means that 1024-core processors will be here by 2015.  

Now throw into the pot Intel’s Multicore Enhanced Layer System, a technology it has developed for stacking multicore processors on top of each other. The quad-core devices are designed with connector pads for this purpose, so unless it proves to be ineffective in practice, we can expect to see it exploited with some relish. Its potential advantage, particularly for server vendors, is that it allows up to four chips to be stacked in a single package, and certainly more in future. 

So with a quad-core chip, systems designers can have eight, 12 or 16-core processor sub-systems. OK, they won’t be cheap, but they won’t be overly difficult to engineer or manufacture – so the price will not be beyond the pale. Now extrapolate the use of this technology with more cores per chip. Many server vendors may well be tempted to have the latest single chip multicore packaged up like this in order to get a temporary performance advantage in the marketplace. 

To return to a well-worn subject, however, server vendors, ISVs and users must first face up to the question of how they get over, round or through the wall that is now facing them. None of this hardware technology will be of any real value to anyone unless they can move the applications they now use to the new environments in ways that properly exploit parallelism, or find new applications that can. 

One thing is certain; most of the applications currently in use will not be suitable.

It may not be immediately apparent to some hardened IT professionals that what happens on the multitudinous desktops of a large enterprise can have a positive effect on the capital expenditure such a company expends on its infrastructure. Even more to the point, I would guess it is a very rare day when the collective `corporate desktops’ are ever seen by IT staff as anything other than a source of pain. 

Yet there are a number of vendors now pushing hard on supplying technology that, at one level at least, seem to be changing the rules on how those multitudinous desktops get integrated with the core enterprise infrastructure, what applications they are delivered and how they are managed. 

For example, what would be the reaction of the average enterprise IT department to the announcement by senior business management that a couple of new offices were to be opened in far-flung, and quite possibly God-forsaken parts of the Globe? I can’t help but assume that, for many, the answer would start with “oh….” And be rapidly followed by the expletive of your choice. Ensuring that those new desktops were fully integrated into the enterprise network could involve some serious engineering, and not a little in capital investment. This would be needed to ensure that there is sufficient processing power in a local server to maintain the operations of those desktop systems, keep communications with base synchronised and manage the applications, and resultant data, running on those desktop systems. And then there is the staffing training to consider….and….and…. 

Yet I have come across one company – currently protective of the competitive advantage it sees itself gaining – that has recently opened three new offices around the world and spent no more in cap-ex than shipping out pre-configured thin clients and an internet connection. These require no more skill to get running and productive than `connect the blue cable to the blue socket’. These new desktop systems are then provisioned and managed from the existing HQ IT resources and require no extra investment in hardware or staff. 

There is an investment in software, in that it uses desktop virtualisation – in this case from VMware. But there are a growing list of alternatives from which to choose, with Microsoft being but the latest (and by no means least) to enter the fray. The trick is that virtualised desktops bypass the conspicuous consumption of PC resources in which desktop applications normally partake, delivering instead just the applications components a user needs to complete the specific task demanded of them.  

The work is faster and more manageable, both for them and the IT department in the background – not least because they don’t have the discontinuities caused by rolling out applications to desktops across the world. 

Getting the world’s enterprise desktops under some sort of control has always been a major objective for enterprise IT managers, but now they can become an integral part of the overall infrastructure, where they belong.

New Mexico in the
USA is not the State that first springs to mind when discussing the locations of leaders in the development of IT infrastructure ideas. But that may just be about to change. The reason can be found at the

New Mexico
Computing
Applications
Center, which is the proud owner of the world’s current third largest supercomputer, which goes by the name of Ecanto, and the plan is that anyone will soon be able to buy time on it. 

In this context, `anyone’ really means large organisations like educational establishments, public authorities and big business rather than you or I. But the essence of this development is that, potentially at least, the early experiments with the concept of `public’ use of both parallel computing architectures and the utility datacentre concept have started. 

Ecanto is, as might be expected, a pretty fearsome beast, sporting some 3,500 or more four-core Xeon processors 28 Terabytes of memory and 172 Terabytes of storage. Such hardware does not come cheap, of course, and so the suggestion follows that
New Mexico is simply looking for way to defray the costs by renting out time on the box. This may indeed be the case, but even if it is true, it only goes to highlight the potential of a utility-based operational model. 

A system like Ecanto will be the very minimum needed for such a task. Indeed it will inevitably need to be more complex if only because the existing system is configured for a different world to that expected in a utility system. To be fair, the Center’s initial targets are educational establishments within
New Mexico, which probably won’t require too much to be changed. But the next target down will be large enterprises looking to run the types of applications for which even they are under-resourced – initially occasional tasks such as complex risk analysis or engineering design. 

The Center’s idea is that the machine becomes, economically at least, self-sustaining and perhaps even a revenue generator, so it will be open to many offers. This does raise an issue, of course, in that Ecanto will need to be adapted and enhanced, mainly in the area of management software, to allow it to cope with a wider workload. But the wider the workload, the better the opportunities. This will also give it scope to offer more comprehensive hosting services than normally available, which in turn should attract more users. 

But by being a supercomputer, and therefore pre-dispositioned to run parallel processing applications, it will be in a great position to offer enterprises a bridge to the introduction of parallelisation in their own operations. My own view is that the inevitable coming of parallel processing, coupled to the growing inability of existing applications architectures to exploit the performance capabilities of multicore processors, will lead to more an more enterprises turning to utility-based services in order to provide those new parallelised applications that will give them real competitive advantage. That way, they will get the advantages of `use’, without the heartache of learning the intricacies of `doing’.  

The institutional supercomputers - and especially the ones, like the Center’s, that are cash-strapped – are in a wonderful position to help enterprises pop a pinkie into the water to test the temperature of parallelisation and utility-based services and experiment with them. Those enterprises that use it to experiment in managing the gap between the new parallel world and today’s essentially serial world could well be the ones that end up with a significant advantage in their respective marketplaces.