Enterprise 2.0

I was reading my Economist on the train this afternoon, and there is a fascinating article on Datacentres (an area I spent some years in ). The online version is here. They were talking about the shifting location of datacentres. Its a nice article, but I didn't like this summary of history:

Until a few years ago, the location of servers was an afterthought, says Jonathan Koomey, a consulting professor of environmental engineering at Stanford University. Most sat in cupboards or under desks. The computers in corporate data centres were often housed in the firm's basement. And dedicated “server farms”, which came of age during the dotcom bubble and often housed the machines of internet start-ups, were mostly built in Silicon Valley and other high-tech hubs.

...so I thought I'd add in some of the background.

There have always been servers in backrooms and garages, and there still are - but serious datacentres even then were more carefully planned than one assumes.

Pre the DotCom era, Telco datacentres were far out of urban areas and put in remote places, and designed so they could almost withstand nuclear winter - some were even under ground. IT companies tended to be in big industrial estates.

Come the dotcom era, Web Hosting companies (anyone recall Exodus?) located initially near big backbone PoPs (points of presence) as access points to the internet was not at that point ubiquitous. Then came the cage-huggers - many dotcoms were so poorly managed (and to be fair, remote ops were much poorer) that they pretty much had to be near their server cages each day, so the demand for web hosting space right next door (and even desks in the datacentre) - Silicon valley, Silicon Alley, Soho in London. This phase too passed as more PoPs appeared, and the industry started to locate around huge telepresence hubs.

And then there was 9/11, and suddenly companies wanted datacentres anywhere except in large cities.

The big issue over the years has been the ever increasing packing of racks and thus the increasing power usage per square foot - and as processors are near-perfect power-wastage devices but most are still designed to work at room temperature, for every watt you put into a data centre, you pretty much have to remove a watt - and as cooling is not nearly 100% efficient, you wind up expending more energy on cooling as on driving. In cities in summer (like New York, or Palo Alto) the added heat can tip systems over the edge. The problem with putting datacentres in the boondocks however is that if you need to run powerlines long distances, the cost and wastage is high

Hence the extreme interest in places with (i) cooler climates and (ii) cheap natural power, that are still close enough to backbone PoPs - and as the Economist notes, those sort of sites are not that common.

Microsoft is looking for a site in Siberia where its data can chill. Iceland has begun to market itself as a prime location for data centres, again for the cool climate, but also because of its abundant geothermal energy. Hitachi Data Systems and Data Islandia, a local company, are planning to build a huge data-storage facility (pictured at top of article). It will be underground, for security and to protect the natural landscape.

Back to the pre dotcom world then - plus ca change

And like warehouse construction before, all sorts of towns, local governments etc will start to give grants to confuse the economic arguments. And like Warehouse towns before them, they will be sadly disappointed by the jobs they bring - a shed is a shed, whether for physical or information inventory.

The issue the Economist does not really mention is the increasing Green Lobby, which will start to also influence where datacentres can go. I also think they will have to start to develop chips that can run at higher temperatures, as every degree higher it can run saves a huge amount of cooling power.

Just seen this on the WSJ:

Hewlett-Packard Co. was close to a deal to acquire Electronic Data Systems Corp. for between $12 billion and $13 billion, according to people familiar with the situation.

The terms of the deal were not immediately clear but an announcement was expected soon, the people said. EDS, which provides consulting and technology outsourcing services, had a market value of about $9.5 billion, based on Friday's closing price of $18.86

Why would HP do this?

A deal would bolster Hewlett-Packard's competitive position versus rival International Business Machines Corp. (IBM) as a provider of services such as tech consulting and customer support. The transaction could spark further large deals in the technology sector as cash rich, mature companies such as Hewlett-Packard look to acquisitions for growth.

Hmm...the system integration and outsourcing industry is a low margin game and gettiing worse as they all chase each other to the bottom line of each big contract. I can see why EDS would want it (27% jump in share price) but does HP really need to buy into even more of a commodity business, and look more like IBM? (EDS makes c 3.5% margin on $20bn, HP makes c 7% margin on $100bn). Isn't the really shape changing play to do something different to the way the rest of the industry is structured, go for growth industries in comms and media instead?

Or is it the only way HP will get anyone to buy their tins

Update - FT reports HP R&D is now going more "D" side, reducing range of projects to the more rapidly commercialised - more signs that current management thinks future success comes from looking exactly the same as the rest of the industry?

Ars Tech on Intel's thoughts re Thin vs Fat clients in a Webservice world:

So what beef does Intel really have against the current Web 2.0, "datacenter plus thin client" model? I think the answer is a bit more nuanced than the simple "they spent $5.5 billion in capex last year and don't want to see the desktop and mobile processor portion of it go to waste" explanation that I started this post with. The fundamental challenge that Intel faces in a computing market that's dominated by massive datacenters is that power constraints will, for the foreseeable future, limit the number of integrated transistors that you can pack into (and therefore sell into) a single datacenter. Even at current transistor counts, most datacenters have large empty regions in them because you just can't get enough power out of the grid to run enough transistors (and, to be fair, hard disks, which are a bigger problem) to fill a modern datacenter.

So datacenter floor space may be theoretically infinite, but the power grid acts as an external economic and physical constraint that places a hard limit on the growth of the market for transistors in any one local datacenter, and this depresses the global market for transistors that datacenters represent.

The solution, then, is to spread the load around by moving as many transistors as possible off of a single point on the power grid (the datacenter) and connect them to all of these other points on the power grid that still have plenty of spare capacity (homes, schools, offices).

This simple, power-optimization and load-balancing problem is why Intel has to keep selling rich clients. The datacenter market alone just doesn't have the capacity to absorb enough 45nm and 32nm transistors to make the massive investment in fabricating them worthwhile, even if you factor in a concomitant proliferation of (low-transistor-count) thin clients. So to have a robust market for transistors at this level of integration you have to keep selling more and more of them across a wider, more broadly distributed swath of the power grid.

There is another interesting conundrum too, where sheer power to the servers is limited to a far lower level than the power available - server boards are superb power wastage devices, ie nearly every Watt of power input to these systems is given off as heat, and because it is so concentrated it has to be removed by cooling systems which are not so efficient (ie to remove the heat requires more energy than the heat contains) - but it all comes off the local Grid where the datacentre is.

We'd add that they're probably on the side of the angels for another reason - since the dawn of the IT network, the Thin Client has emerged every so often as the "obvious win", but it always seems to die away again as (i) people seem to vote time and again for some independence from the central App, and (ii) the network is never "always here, always on" - and at some point it fails you, and you swear "never again".