Clive Longbottom: Is it all just a matter of scale?
Tue 8 Apr 2014

As virtualisation and cloud computing begin to creep towards more general usage, the approach seems to be built on the premise of the use of a scale-out approach built on a mass of basic Intel-based blade (or similar) servers. This “commoditisation of IT” has been much written about – but is it really true asks Clive Longbottom?
For some workloads, just throwing more CPU resources at them will be enough. These workloads will not be massively network intensive or disk intensive: number crunching will be where the problem lies. As long as the application is written accordingly, a base of virtualised Intel-architected servers will hit the spot.
However, commodity scale-out doesn’t suit every workload. Increasingly, the Intel vendors are realising this and are engineering systems so as to be more akin to scale-up machines. Cisco’s UCS, Dell’s Active Systems and VCE’s Vblock are all instances of systems that are built to converge the server with storage and network components to provide a better platform for more “blended” workloads.
However, this still leaves those outer worlds of non-Intel systems. Although the death of the mainframe has been touted since the PC first hit the desk in 1981, it is still not only present within many organisations, but continues to sell well for IBM in hardware and software revenues. Why? Because it is a good platform for certain workloads. It does a great job with its native zOS in dealing with on-line financial transactions; with zLinux, it deals well with thousands of concurrent virtual machines.
IBM also has the Power chip. Having sold off its x86 activities to Lenovo (although strategic work will still carry on between the two companies), IBM will be putting more focus on the capabilities of its Power-based PureFlex and Power systems as a platform not only for specific workload types but also to compete alongside many Intel-based platforms.
Oracle is busy revamping its UltraSparc/Solaris range of systems. Although unlikely to compete strongly in the general server market, Oracle can make a strong play in specific areas (such as the telco space) where Sun had a strong presence and a complete stack from hardware to application could make a solid proposition for a company seeking simplicity in contracting and systems management.
Although the increased use of virtualisation and cloud computing is leading many to think that Intel (or AMD) is the only game in town, in fact it should be seen from a different viewpoint. The whole idea of virtualisation is to abstract the upper software layers of the platform from the hardware layers. Therefore, heterogeneity should be far easier than it used to be – and the right workloads can be placed on the right underlying resources.
Yes – this requires intelligence in the software layers: workload placement needs to be automated as much as possible. Again, IBM understands this, and with its capabilities built up through its experience with platforms like the zEnterprise, it is close to being able to direct workloads to the right platform.
Indeed, such use of virtualisation and intelligent workload management should usher in increased use of offload engines – such as Atom-based micro-servers, NVidia GPU systems, Azul Systems Java machines – leading to hyper-tuneable, hyper-scalable platforms.
The stack is getting more interesting – don’t assume it is commoditising.