Google’s server requirements are such that the company resorts to building its own systems, and given the shear scale of its operations it makes a great deal of sense. It can design and specify systems that exactly match its service delivery requirements, rather than having to compromise around what the vendors are willing or able to offer. There is an infrastructure lesson for enterprise managements to learn here– both business and IT - and it may not be the obvious one. 

That obvious one is the assumption that other enterprises should follow Google’s lead and start building their own, though there may be a few for whom such a route does make some sense. After all, if a business does need several thousand servers a year, all of a particular specification, it does make economic sense. For example, Google can use low-grade memory chips in its servers because, while it needs a huge amount of memory, it is not important that it is the fastest ever made. Shear volume beats brute speed in practice. The same goes for storage: huge amounts of cheap slow(ish) disk storage is better value, both economically and for service delivery, than fewer fast, expensive ones. 

But those are rules that will not fit the majority of enterprise users, so the economics do not make sense. There is an argument that there might be sufficient mileage for the vendors in making low-spec servers with a very aggressive price, but even then it is unlikely that there will be sufficient demand for a single specification to make it workable – too many enterprises will find the selected spec just too far away from what they really need. 

The lesson from Google, therefore, is not aimed directly at the enterprise users themselves. Instead it is aimed at the server vendors. It is common for pundits to suggest that the outsourced utility datacentre model is still years away from being available. It also common for them to say that Software as a Service is something of a specialist niche for specific functions like CRM. In answer, I would just suggest that Google – with Microsoft going gang-busters to catch up having realised the market dynamics are pointing away from it and towards Google – are already being very successful offering outsourced datacentre services.  

Every Google search is an access to an outsourced datacentre offering a SaaS delivery model. And more and more enterprises are signing up with Google for just such dedicated services, without ever once getting paranoid or protective about ` their data’. It is the way we are all heading, without actually realising it. 

And that – building contenders in the outsourced utility datacentre market- is the opportunity now facing the server vendors. The big ones have the clout to play an interesting financial game here – selling servers to themselves. This means they can cut the unit price margins to the bone because the revenue will come from the services they offer. The likes of IBM and HP already offer outsourced datacentre services in a small way, but the market is now just coming ripe to do it properly – Google is demonstrating that.  

Such a model also makes sense of the users’ needs for differently spec’d servers. The vendors can build a reasonably spec’d unit in huge volumes and then couple them with the management software they have that allows individually spec’d servers to be virtualised out of the primeval soup of the total resource. Users can get what they want – nearly always a great deal faster than by ordering it – and can get rid of it when they no longer need it: click the `thanks, now go away’ button on-screen and pooooof, the server is no more.  

Greater flexibility and agility for users, coupled with better cost management, is matched by better predictability on server demand for the vendors, coupled with better cost management. And it is not a pipe-dream for the future. Google is doing this now, and is not failing in the attempt, either. 

It was, of course, eye-catching to describe IBM’s new Mach-Zehnder electro-optic modulator technology as the way to get supercomputer power into a laptop. But it was also rubbish as well, if we’re all honest. 

First of all, the average laptop will always have the equivalent compute power of a supercomputer of some period in history, usually not that long passed either. Second of all, such a comparison is rather pointless because most supercomputers are parallel processing brutes running complex floating point calculations on vast arrays of data – not even remotely close to the jobs an average laptop has to perform.  

But imagine the potential of the new modulator being used in a datacentre – a very large datacentre – the impact could be dramatic.  

Just a quick recap to start with: the modulator is a new IBM technology, very much at the early development stage. This takes electro-optics to new levels of micro-miniaturisation allowing light to be used as the signal transmission medium within a complex processor chip, rather than electrical signals. Just to avoid going over-board, it is fair to point out that it may not make it as a workable technology – like IBM’s superconducting devices before it, the problems associated with its practical application may prove such that the lyrical hyperbole of its potential may yet be consigned to the bin. 

But back to some lyrical hyperbole. The advantages of using light for internal processor communications means several things: reduced energy consumption and heat production are two very important ones. Given that a typical datacentre rack now consumes around 12 kW of energy and uses a further 12 kW to provide the air-conditioning needed to keep it working, any saving there is good. This could be a route to significant savings.  

The use of light also greatly increases the bandwidth of the processor, meaning far more work can be done. Not only does a single light source offer better bandwidth than the electrical equivalent, but it also means that multiple light sources, using different colours, can be used in the same physical channel, multiplying the bandwidth even more. Couple this with a real multi-core processor design that mimics the massively parallel processor architectures of supercomputers – several hundred cores per chip at least – and some real advances in not only computer design, but software design as well, will be possible. 

And who will want blindingly fast, dense, cool (OK, cooler) datacentres using tolerable amounts of energy to provide real high workload performance? Well everyone, of course, except that most will feel certain even now that they will not be able to afford it. But then again, a datacentre like this – or a good few of them scattered around - will mean only the most lavishly endowed few will even consider the purchase option. 

Most will be outsourcing, either through a managed services operation or a SaaS provider. And a technology like this is just what is needed to make such outsourcing, ultimately, the only sensible option for the majority of users.