• Support Analyst Programmer
• Business Program Manager - Large...
• Technical Consultant / Analyst
• C++ Developer - Melbourne - 50...
• Senior Desktop...

Down the technology track in 2008

Information Age staff, Information Age

04/01/2008 04:29:53

Shuffling virtual servers for optimum power efficiency

By Ted Samson

Today's data centres squander AC power on excess capacity. But what if you could assemble dynamic, virtual server farms on the fly and consume only the power you need?

It's a pleasant afternoon in NSW (I'm pretty sure that's happened before). You're a data centre operator walking the floor of your facility just outside Sydney -- but it's eerily quiet, considering that fleets of servers are running all your mission-critical applications.

Once upon a time, the absence of humming might have meant something was catastrophically wrong. But a glance at your systems' management UI confirms that, indeed, your apps are being served. It's just that your servers in your India-based data centre facility are doing most of the work -- and conveniently, it's off-peak hours there, which means you're paying a fraction of the energy costs.

The fact that machines are self-adjusting their power consumption and virtual workloads -- even powering on and off as needed (or not) -- helps shrink the electric bill even more.

Energy efficiency, one of the key concepts under the umbrella of green technology, has become increasingly important to data centre operators who are struggling with soaring electric and cooling bills, not to mention limited power from local utilities and space in existing datacentre facilities. From the chip level up to the software layer, vendors are devising ways to reduce the amount of juice given machines while wringing as much work out of each box as possible.

And based on what we're seeing, I expect the next big thing in green tech will be what I'll dub the dynamic server farm, a hybrid of power management, systems management, Web services, and virtualisation technologies.

One piece of this puzzle: power management at the hardware level, which we're seeing crop up in systems management tools. Hewlett-Packard, for example, recently unveiled a new power-capping feature in its Systems Insight Manager hardware management platform. The capability lets admins control, on an individual machine level, how much power a given server consumes.

As HP puts it, while a hardware vendor might tell you that your system needs 1000 watts of juice, that's only if the machines are running at 100 percent capacity. But best practices in the datacentre dictate that you not run a machine at that speed; if you do, you're just begging for trouble should demand further increase.

Now, what happens if all your servers are running at a safe 80 percent utilisation threshold and there's suddenly a spike in demand? You've got management software in place that will simply wake up other servers that were cosily and inexpensively snoozing since they weren't being called upon to act.

Recently, Appistry introduced that very technology, called EnergySaver, to its Enterprise Application Fabric platform.

In a similar vein, VMware has spoken to me about plans to incorporate power management features to its server virtualisation product line. The way it would work is, it would dynamically transfer underutilised VMs to as few servers as possible during downtimes and power down the servers not in use.

(Virtualisation is, of course, another key component in this dynamic server farm, as it's the poster child for energy efficiency.)

So now we've got individual servers consuming only as much energy as they need to do their job. We've got servers waking up only when they're needed -- and going back to sleep when they're not. And we've got virtual machines in play, which in and of themselves yield more bang for your buck from hardware, only they're being dynamically moved among servers to ensure that as little power as possible is being used.

But what if those VMs could be dynamically moved beyond the confines of a given data centre to, say, one in a different time zone -- based, at least in part, on both energy supply as well as energy costs at a given time? Your local utility sends out an alert, as a Web service, that there's a demand spike in your primary data centre, rates are about to double, and a brownout is imminent.

But your dynamic server farm management platform knows that it's already after-hours in NZ, where energy prices are lower and supply is ample, so it dynamically pushes the server load to your facility in Auckland.

And if Auckland, why not to your facility in Bangalore or Shanghai or wherever the energy prices are lowest -- as long as you can ensure sufficient service levels? (Your dynamic server farm platform would, of course, be measuring service levels continually.)

It's a pretty ambitious green technology vision, but if you think about it, many of the technology pieces are already out there, and just as important, so is the demand. Technologists often think of data as the lifeblood of IT, but data doesn't fuel your equipment -- energy does.

Storage: Services in the cloud will meet tomorrow's storage needs

Storage remains a bear to manage -- and few signs point to significant improvement. The solution? Let someone else do it

By Mario Apicella

Despite vendor promises, managing storage is getting harder, not easier. Storage vendors boast of smart applications that ease storage management to the point that a caveman could do it. They promise tools that discover neighboring devices and servers without human assistance - and that automatically configure themselves accordingly.

As is often the case with sales pitches, those claims are partly true (or only half lies). Make a small detour from the standard configuration, and the easy setup promised by such tools fails miserably, leaving inexperienced admins to fend for themselves.

Even the much-touted simplicity of iSCSI SANs has failed to materialise. For a novice, the advantage of iSCSI over FC (Fibre Channel) boils down to not having to learn another class of switches and networking gear. But getting iSCSI hosts and target devices to work together requires skills that were never needed to manage a LAN. To add insult to injury, most storage management suites completely ignore iSCSI. And new technologies on the horizon, such as FCoE (Fibre Channel over Ethernet), will only add complexity to the administrative puzzle.

How will customers cope with the conflicting needs for increased capacity and cost effectiveness? For many, SAAS (storage as a service) will be the answer: Instead of paying through the nose for storage devices and their maintenance, CIOs will find it more convenient to rent storage space - and divert the money and resources saved to more strategic projects.

Initially, SAAS will be sparked by faster and more affordable bandwidth, along with the surplus capacity of titans Amazon, Google, and the like. Other vendors will follow, creating Internet-based storage services of unprecedented size that will undercut that market domination of such companies as EMC, HP, and IBM.

SAAS will make storage more affordable and granular - you pay for what you use and no more. Best of all, it will empower customers to deploy technologies whose storage costs would have been otherwise prohibitive.

Replacing in-house apps with applications in the cloud

By Oliver Rist

Applications delivered through the browser keep SMEs nimble by eliminating application maintenance costs

In the near future, there's only one way to go for SMEs when it comes to purchasing business software -- and that's out of house. Whether it's full-on SaaS (software as a service), where users access all facets of the application through a browser, or a hosted product (including hosted Exchange, where only the server component is off-site and users employ a standard desktop client such as Outlook), either model is simply too cost-effective for SMEs to ignore.

The only variable unaccounted for in this model is that of the Internet itself. Many SMEs simply don't trust the Internet as a mission-critical delivery model. But that's changing. The number of business-disrupting Internet outages reported in most metro areas has dropped precipitously over the last five years, and those stats are only going to get better. Additionally, Internet backup services are offering more options for SMEs at increasingly reasonable rates.

Anyway, a little redundancy comes cheap these days. Whether it's T1 service backed up by DSL or DSL backed up by business cable, it's pretty inexpensive to keep two Internet connections running -- especially when you subtract the salary for a full-time IT person to maintain in-house apps because you've opted for a full suite of equivalent SaaS apps run entirely by off-site providers.

While this may not be possible for businesses that need specialised software, those able to run on mostly standard packages can do so with lower installation and TCO numbers, no permanent on-site administration staff, and a more in-depth feature set than they'd likely be able to set up themselves. And they can do that today.

With the growing popularity of Web 2.0 technologies and application frameworks being built on top of existing online hosting services, such as those from Salesforce.com, even SMEs with specialised software needs will soon be able to meet those requirements online. Large enterprises may have different concerns, but for lean, mean SMEs looking for maximum features and flexibility with a minimum overall cost, SaaS and hosted software simply have nowhere to go but up.

Databases will pack larger sizes into smaller spaces

By Sean McCown

Databases are ballooning to gargantuan proportions, killing performance. That's why the next frontier is new compression technology

One of the bigger problems databases face is the space crunch. As individual databases edge toward the petabyte range, it's getting much harder to find appropriate storage. And of course, the number of tables isn't growing in proportion, which means we're going from a large table size of a few million rows to that of a few billion rows. In fact, soon enough, few among us will be surprised to encounter tables in the tens of billions of rows.

Not to mention that storing all this new data is only a third of the problem. Making it accessible is the second third: Sure, disk space keeps getting cheaper, but coaxing decent performance out of a database in the petabyte range may require thousands of drives.

The last third of the problem is simply the space required to maintain backups of all this data. Currently, it's already too expensive to back up a database that's several terabytes -- without compressing it, that is.

That's why I feel the next big achievement in databases will be better, more efficient compression algorithms and surrounding structures for table data -- and possibly even for databases as a whole. The compression technologies currently in use are neither widespread nor high-performance enough to withstand the rigours of a demanding DSS (decision support system) or OLTP (online transaction process) system.

As for backups, a few vendors currently sell compression solutions for SQL Server, but no solution of much consequence exists for the other relational DBMSes. And all of them rely on the usual open APIs and standard compression algorithms. To get the level of compression and performance that the future of databases will demand, there's going to have to be a technology breakthrough.

That's why I believe that live data compression and backup compression will be the next big frontiers to conquer.

Sharing behaviour across a service-oriented landscape

By Dave Linthicum

New middleware will expose methods or service calls as if the whole application infrastructure were running on one big, local system

Middleware is morphing from something that only deals with the movement of information from one place to another to something that's much more sophisticated and useful in today's emerging service-oriented world. The emphasis will shift from machine-to-machine integration to the management of true services or behaviours.

What's the difference? If you think of middleware as a pipe, right now most of what flows down that pipe is simple information -- such as customer data, sales data, and so on -- all bound to messages that are similar to small database records moving from system to system.

Of course, middleware has become much more helpful over the last 10 years, adding such features as transformation, reconciliation of application semantics, and in some cases, the assurance that information flows to the correct machine.

But the movement toward service-oriented middleware is really all about dealing with functional behaviour -- and the way in which that behaviour is shared among systems. In other words, instead of just sharing information, the middleware facilitates cross-invocation of remote methods or service calls as if they were local. Naturally, data is bound to these methods, and as with traditional objects, you leverage the methods to control access to data.

So why is this new? Even with the use of such middleware as integration servers, message brokers, or service-enabled queuing systems (ESBs, for example), the focus has remained on the exchange of information instead of services. Today, information-oriented middleware is currently being deployed within SOAs for service-oriented purposes, with limited success. Driven by the need to support service orientation, a new emphasis on behaviour-sharing will fundamentally change how middleware is built and deployed.

Security: The great privacy compromise

By Roger A Grimes

No magic technology bullet will solve the ongoing crisis in enterprise security. The answer is political -- and a long way off

Security will remain at its current pathetic level or worse during the next five to 10 years. The next big thing is a long way off, primarily for cultural rather than technical reasons.

Until then, I can confidently predict that every new security product introduced to protect you will fail miserably. Criminals will continue to hack at will and almost never get caught. Whatever defences are envisioned and deployed, malicious exploits will bypass them. It's not a pretty picture, but it's a pretty sure bet.

At some point, with way too much pain and blood on the ground, we will decide to solve the real problem: the pervasiveness of anonymity. Think of almost any Internet security problem (outside of encryption for confidentiality) and default, persuasive authentication would fix it. There'd be better authentication of our computers, hardware, booting, OS, and applications, as well as authentication and identification of who sent what network packet, from beginning to end.

As for end-user authentication, biometrics will finally gain critical mass. For passwords to be secure, they must be complex and eight or more characters long. But soon eight characters won't be enough, and it will take 10 complex characters to be relatively secure, with more frequent password changes.

That's why biometrics and two-factor solutions will become standard in the corporate environment over the next five years. Online banks, stock trading sites, and other financial institutions will require two-factor tokens. Most governments will mandate biometric identification to obtain services.

Computers and all network devices will become better identified using persuasive trust mechanisms. Receivers of network packets will be able to trace all network packets and connections (egress and ingress) back to their origination.

Privacy advocates will protest the default authentication built into everything using a computer chip. But the benefits promised by corporations and the government ("We can help you locate your children if they are kidnapped") will make consumers beg for the intrusion.

Privacy advocates who don't wish to be "chipped" will use their own version of the Internet and take their chances in a Wild West-style environment much like the one we suffer today.

Meanwhile, in the safe confines of an Internet transformed by pervasive authentication, malicious hackers will have a hard time escaping capture. The increased protections provided by more secure identity and authentication mechanisms will make malicious hacking too arduous to be profitable any more. In exchange for compromising on privacy, the online experience will finally be a safe one.

Networks: a fabric smarter than its end points

By Paul Venezia

As switches, routers, and servers bulk up on processing power, the pendulum is swinging back to intelligence residing in the network itself

The Internet was built on only a few rules, but one of them was "keep the network dumb and the end points smart". That means that networks are designed to know little about what they're carrying, pushing the onus of consistency checking and packet handling to the server and the clients.

That architecture has served the Internet and private networks very well in the roughly 30 years since the advent of TCP/IP. In the next 30, it's going to change -- big time.

Back in those days, there were many good reasons for networks to remain dumb. The state of networking technology and processor development were the major factors. The erstwhile Ethernet hub, for instance, was perhaps the simplest network device, and easily the least secure. Routers ran very limited code on very limited hardware, and the thought of deep-packet inspection wasn't even a pipe dream.

Gradually, the network has become smarter: quality of service, firewalls, and layer-3 switches have all enhanced the security and speeds of local and wide-area networking. Today, the rapid pace of development in solid-state storage and low-power, high-speed CPUs is truly changing the game.

It's possible to inspect every packet between gigabit interfaces and make intelligent decisions on whether the communication should take place or what direction the traffic should be routed -- all at wire speed.

Next up: networks will become smarter than the client. Advances in virtualization and thin-client technology are reducing the processing power present at the client side, just as switches and routers are gaining smarts and speed. More and more server tasks will move into the network space, running on core switches.

Desktop systems will disappear in favour of monitors with an Ethernet connection, and servers will return to their mainframe roots, offering little more than storage and a hypervisor layer for a few applications. After all, what do you think made George Orwell's telescreens so ubiquitous?

Desktop technology: streaming to a screen near you

By Randall C Kennedy

The technology is finally here for application virtualization, which solves the desktop management problem that has cursed IT for decades

Shrink-wrap is dead. The era of the traditional end-user seat licence is coming to a close. Tomorrow's enterprise desktop is all about subscription-based delivery of customized computing stacks via application virtualization.

Of course, we pundits have been saying the same thing for years. But this time, the technology has finally caught up with the vision. With Vista and Office 2007, we may well be looking at the last hurrah for Microsoft's classic desktop licensing scheme. Microsoft's acquisition of Softricity -- and its SoftGrid application virtualization platform -- provides the Redmond behemoth with the missing piece in its long-term, subscriptions-based computing strategy.

With SoftGrid, Microsoft can virtualise and stream its entire lineup of rich-client applications. This includes Microsoft Office, Microsoft Dynamics, Microsoft Games -- the works. These applications can be delivered in all their rich-client glory right over the Web, without any of those messy licensing or configuration management issues.

It's all thanks to the magic of application virtualisation, which, over time, will prove so effective that Microsoft will quickly reposition SoftGrid as the preferred (if not the only) way the company delivers applications. Period.

So hang on to those old Office 2003 and 2007 installation CDs. They'll fetch a pretty penny someday on eBay.

The real loser in all of this is VMware. Its vision of a virtualised appliance-based future will go down in history -- alongside Windows Terminals and those awkward Windows CE-based hybrid "notebooks" -- as yet another attempt to fill a round hole (configuration management) with a square peg (virtual machines).

Note to VMware: Virtual desktops were a bad idea 10 years ago, and they're an even worse idea today. Do us all a favour and stick to the data centre. That way, when the Linux GPL purists finally catch up with you (and you know they will), we'll know just where to find you.

Virtualisation: a new era of roaming desktops

By David Marshall

Application virtualisation will make IT's life much simpler by separating application code from the platform on which it runs. And the technology is ready to roll

When most people talk about virtualisation, they mean server virtualisation. But you don't need to look any further than VMware's IPO to realise that particular train has already left the station. When Wall Street gets it, it's no longer the Next Big Thing.

The really exciting virtualisation technology -- in fact, the most exciting technology to hit ICT in a long time -- is application virtualisation. It hasn't been widely deployed yet, but it's poised to take off.

Microsoft and Symantec have seen the potential: The former acquired Softricity SoftGrid, while the latter purchased Altiris SVS. But many other players have entered this emerging market, including Citrix, DataSynapse, Thinstall, and Trigence, to name a few. And the startup activity in this space is still full of life.

Each application virtualisation vendor takes a slightly different approach, but the objective is the same: to separate application code from restrictions imposed by individual servers, operating systems, and clients, allowing the application to run regardless of its environment. Application virtualisation liberates the application from the operating system much like server virtualisation has liberated the operating system from the underlying hardware.

One immediate effect is that the complexities of upgrading and patching are eliminated. Because only one instance of the installed application exists for many clients, only one instance needs to be upgraded or patched. The process is easier, faster, and far cheaper to administer.

IT departments not only can see what applications are in use, they can also manage and control the applications to stay within licence compliance. And it's in the best interest of software vendors to create their applications so they can be packaged in the correct format to be properly virtualised.

Longer term, the technology will foster more separation between user data and the underlying application itself. In essence, users will enjoy a desktop environment that follows them around.

And best of all, application virtualisation will be married to streaming technology to deliver both the application and the "roaming" desktop to the end-user instantaneously, whether he or she works in the office or remotely. Application virtualisation solutions will bundle security, licence management, and the ability to revert the application back to a pristine state -- or turn the application on or off from a central location.

The economies of scale are mind boggling. Can anyone imagine a technology with a more direct impact on IT workload and software costs?

Business models: a million ways to pay as you go

By David L Margulius

In a few years time, what will it be like to buy software by subscription? Consult your mobile phone calling plan for a preview

The big thing in business models is the subscription, which will be a winner for IT vendors and their customers alike for the next decade. No doubt you've heard that one before. But there's a twist.

First, a little history: traditional business models have been in flux over the past decade in every corner of the global economy. The Internet has given buyers more control and transparency, and levelled the playing field for competition. In IT, packaged software is facing strong challenges from both services (consulting bundled with software) and open source (support plus free software).

But subscriptions have the mojo. They started in the days of Ben Franklin, essentially as a prepay that helped publishers mitigate the risk of both fixed and variable printing costs (presses, paper, ink, binding). Then came telephones, cable TV, and Netflix.

Next, high-speed Internet access lowered the variable costs of deploying and integrating software, enabling companies such as Salesforce.com to prove the viability of SaaS (software as a service). In the meantime, customers have grown increasingly unhappy with conventional software licensing and deployment, which requires big money up front for an on-site install -- and lets vendors lock in customers and milk them (also known as the "razors and blades" or "drug dealer" model).

The lower switching costs of a subscription plan empower customers to hold vendors to account and demand ongoing value.

So what's the twist? Subscriptions will become much more complex and specialized, just like consumer loyalty programs and financial products. What now sounds simple ($75 a month per seat) will be sliced and diced every which way to cater to every imaginable need and customer segment. Loyalty will be rewarded. Service levels will be segmented. Options and futures will be baked in (forward and backward road map compatibility, and so on).

The good news: even if subscriptions become as complex as licensing deals, they'll still be more transparent -- you'll know exactly what you're paying for as you order from an a la carte menu. The days of the golfing, boozing sales guy negotiating a price "just for you" are over.

The winning model dispenses with that guy and puts his salary into product development, leverageable across many more customers paying standard subscription rates.

No more hidden fees, gotchas, or endless negotiations. It'll be Henry Ford all over again: Value for value received in nice, standardised bites. Where do I sign?

High-performance computing: supercomputing everywhere

By John West

Plummeting costs, Moore's Law, and creative new applications are making high-performance computing a commodity

Supercomputing isn't just for spies and nuclear scientists any more. In fact, supercomputers are probably going to play an increasing role in your business's bottom line, starting right now.

Governments and large science and engineering organisations have used supercomputing (also known as HPC, or high-performance computing) for decades to solve the complex equations that describe the physical world.

Supercomputers can predict how well a new car design will survive an impact before a prototype is even built, or they can help geologists predict the best way to retrieve oil from a new well.

But HPC offers real benefits in a variety of enterprises, and even the most old-fashioned companies are starting to adopt supercomputers to run core business operations. Manufacturing facilities, for example, are using HPC to compute optimum flow conditions for moulding parts, avoiding costly trial and error. And delivery companies are using supercomputers to calculate the most efficient routes, saving time and fuel.

The shift to building supercomputers with commodity rather than custom processors a decade ago has created an extensive body of knowledge about how to craft supercomputers from commonly available parts. For these machines, price scales with size.

The same technology that builds supercomputers for governments can now be used to build 100-processor machines for just about anybody with $10,000. And companies such as HP and Dell are selling preconfigured supercomputers on the Web.

Your regional delivery company may have only 256 processors now, but as new opportunities emerge to improve processes from payroll to personnel management, it will need ever larger supercomputers. In the not-too-distant future, the world's largest computers may be as likely to serve your grocery store chain as they do governments and multinational energy companies.

Open source: Innovation through recombination

By Dave Dargo

The stage is set for open source to reinvent itself, fashioning new solutions by borrowing the best code across a broad spectrum of software

Today's software business model isn't about writing software; it's about the enablement and servicing of software. And that just happens to be what the open source movement is all about.

So far, the open source movement has tended to develop software that mimics commercial products. We will soon see it graduate to creating entirely new solutions. One of the great freedoms of open source is the ability to take just the bits we need and recombine them into new applications, accelerating innovation and time to market.

The first change we'll see is the elimination of the general purpose operating system for servers. We'll no longer need to drag around senseless OS components with their broader security and availability vulnerabilities when we just want to deploy a database server, an application server, or a Web server. Linux is a great OS, and it's an even greater repository of discrete OS components.

Open source allows startups to take the bits of Linux and other open source projects they need and combine them in new ways that deliver new capabilities. Companies have already started down the path of designing configurable Linux configurations meant to be combined with other software components to create purpose-built solutions. Open source allows us to move beyond the stale traditions of certified stacks of software to truly integrated solutions that home in on specific needs.

Open source licensing promotes this type of innovation, and the market will ultimately reject proprietary licensing models that prevent it. The freedom to innovate is the greatest freedom of all -- and the perfect building material, in the form of existing open source code, will accelerate it to unprecedented levels.

App development: the new face of Internet apps

By Martin Heller

Blending the best of conventional desktop and Web apps, rich Internet applications occupy the hottest app dev real estate around

A huge chasm divides the application development world, one that dozens of vendors are rushing to fill. On one side are desktop applications; on the other are Web applications. In between are RIAs (rich Internet applications) which have attracted droves of developers and will continue to do so for years to come.

Desktop applications respond quickly, present complex user interfaces gracefully, and take full advantage of the resources of the local computer. On the flip side, they require an installation procedure, must be compatible with all the other locally installed software, and need to be updated, which may introduce incompatibilities.

Classic Web applications need no installation, are always up to date, and can be made universally compatible. On the other hand, they tend to be unresponsive, their user interfaces are limited, and they can be knocked for a loop by server loads and connectivity issues.

RIAs attempt to combine the strengths of desktop and Web applications without falling prey to their weaknesses. RIAs try to present most of their user interfaces at the client so that they can be responsive and the interface can be as complex as it needs to be. RIAs often do need an installation, but usually only for the runtime engine, which tends to be small and most often updates itself automatically. The RIA application itself typically executes on the remote server.

RIAs try to allocate resources to the most appropriate place. If the gating issue is the overall scalability of the application, then the designer of the RIA will run most of the CPU-intensive computations on the client. On the other hand, if the application uses a database intensively, then many actions will run on the server.

Many RIAs are written to accommodate intermittent connectivity. If such an app needs a database resource, a local database kicks in when the local computer disconnects from the Internet. When an Internet connection is re-established, the application synchronises the local database with the central database.

I don't believe for a minute that RIAs are going to take over the world. There will always be applications that are best implemented on the desktop and others that are best implemented on the Web. What I believe is that RIAs will play an increasingly important role in bridging the gap between desktop and Web applications, and eventually at least one technology may come along to unify the whole spectrum.

I'm not sure what that technology will be -- or even whether it's something we've already seen or has yet to be created. But I expect for it to emerge in the next five years.

Processors: dividing chips into many virtual cores

By Tom Yager

The era of the big, muscular CPU is on the decline. The real trend is toward slicing and dicing cores virtually, one for each thread

The current approach taken by x86 CPUs -- to stuff as many processor cores and as much cache memory as will fit on one chip -- will prove impossible to scale beyond a certain point. And adding more, big, hot processor cores may not be the best fit for server roles that call for managing large workloads over long periods of time.

Over the next several years, I believe that AMD, IBM, Intel and Sun will gather around an objective epitomised by Sun's UltraSPARC T2 (Niagara 2) CPU: an approach to maximising throughput that Sun refers to as CMT (Chip Multi Threading).

The x86 world has seen the likes of chip multi-threading on a smaller scale with Intel's Pentium 4 (Netburst) architecture. Intel's Hyper Threading split one physical CPU into two logical processors, but it was implemented on a complex and unwieldy processor architecture.

Intel shelved Hyper Threading when it scrapped Netburst in favour of Core micro-architecture, but Intel's move to a simpler CPU core paves the way for Hyper Threading's return. To actually implement multiple threads, however, Intel may have to forgo its obsession with equipping each new generation of CPU with more cores, bigger caches and faster clock speeds.

It could take years, but Intel and AMD will go wide on threading. All CPU makers will, because unlike a processor core, which has to be wired into cache, memory, and I/O, hardware threading splits the gross resources provided by cores in a way that suits virtualisation particularly well.

And unlike virtualisation extensions, which require that system software be aware of the means by which the CPU supports multiple virtual partitions, chip multi-threading doesn't require anything other than whatever is already programmed into OSes and virtualisation solutions to deal with multiple discrete processors.

In the end, the biggest gains in server throughput will come from taking more of that work away from software.