Jackson Pollock was the father of “action painting”, dripping paint on to canvas and even using jet engines to create his abstract art, now worth millions on the art market. Although he used the same basic resources as any other painter (oils and canvas), he used different techniques to create something that is pretty easily discernible as being a Pollock.
With IT, there is an increasing move toward abstraction as well, with the hardware and software layers split apart using software-defined constructs. In essence, the idea is to use the same resources as everyone else, but differentiate through the software-defined environment to create something that acts and behaves differently according to needs.
Does it work quite so easily, though? To put it quite simply, no.
“Software-defined” is a great way of commoditizing certain tasks such as systems management and dealing with certain non-time critical data. However, where time criticality is involved, software-defined environments can start to introduce additional latency, as packets are shifted from the underlying physical environment through the virtual layers into the abstracted software environment, and then back again.
The impact that this can have on the physical environment cannot be overlooked. In the early days of software-defined networking (SDN), much was said about how completely dumb boxes would take over, as all the intelligence would be carried out through the software-defined control and management planes. However, what we are seeing is a combination of SDN (along with network function virtualization (NFV) in many cases) and hardware-based intelligence that ensures that the optimal overall performance is maintained.
The same is happening at the server level. Although for many, a single scale-out platform of Intel-based servers can offer a suitable platform, more organizations are finding that what a software-defined environment offers them is the capability to go for more heterogeneity, using a mix of standard, ‘commodity’ Intel/AMD servers in conjunction with graphics processing unit (GPU) based engines from the likes of Nvidia or Supermicro, specialist offload engines such as Java workload servers offered by Azul Systems, Power-based IBM systems, or even mainframes. These physical systems can be hypertuned to manage specific compute workloads, should these be on-line transactional processing (OLTP), big data, communications, or other types of workload.
The same has been happening in storage — mixed workload systems are now coming through from the likes of EMC, IBM, and others, with software in conjunction with intelligent array controllers being the arbiter of what goes where, when.
This will then drive a need for the One “Software-defined” To Rule Them All — trying to maintain a technical platform with a software-defined network approach from one company, a software-defined storage approach from another, and software-defined compute from yet another will fail horrendously. Even using standards in a siloed approach for each area will not work.
EMC’s software-defined data center (SDDC) was the first shot at bringing everything together under a unifying system. However, it is still early days, and more will be required to ensure that an overall platform — which will consist of some physical systems, some virtualized ones, and private and public cloud services — will all work in harmony.
To me, a Pollock painting looks a bit of a mess (somewhat akin to many existing technical platforms I see in some of the organizations I talk to) — the artist essentially left where the paint ended up more to chance than anything else (whether it be through liquid painting where gravity was the main force, or his action paintings using jet engines where little overall control could be put in place by the painter). You really do not want to end up with a complete Pollocks of an IT platform: control is required at granular levels in context with everything else. However, by taking a long-term SDDC approach to overlay a heterogeneous mix of the right hardware required to give an optimal platform, the chances are that you could end up with an Old Master of blinding beauty — rather than a collection of smears on a canvas.
Image credit: Bill Stilwell (flickr) / CC-BY-SA