WavebreakmediaMicro - Fotolia
What will the data center look like in five years? The answer is complicated by the current surge of innovation in almost every sector of our industry.
From 2015 to 2020, we'll see the highest rate of change in almost every area, excluding dinosaur legacy gear and much of what makes networks, servers and storage tick will alter drastically in that timeframe.
It's worth looking at the core technologies first. Changes here redefine the meanings of networking, servers and storage.
A current trend, and possible fad, is the convergence or hyper-convergence of servers and storage into a single box. Driving this is the availability of super-fast non-volatile memory express (NVMe) solid-state drives (SSD), which saturate their hosts with as few as six drives. That's a good amount for a server box, and so the idea of just parking the drives in servers and not using separate networked storage boxes seems to make sense.
Software such as VMware's virtual storage area network (vSAN) lacks in performance, perhaps due to under-optimized I/O paths, leaving room for additional motherboard or other server optimization and for better local area network connectivity in data centers of the future. However, the latter issue will resolve itself within a couple of years since the Ethernet is exploding in performance and adding RDMA capability.
The server motherboard is a whole different scale of change. Everything is on the table, from tight coupling of dynamic random access memory (DRAM) and CPUs to real-time memory compression. Persistent memory is closing in on DRAM, with Intel planning 3D XPoint DIMM-based offerings at roughly 2.5 times the latency of DRAM, as well as using the 3D XPoint as an NVMe NAND drive alternative, where XPoint will be roughly 10 times faster than these already fast drives. Assuming pricing is realistic, XPoint will change the structure of top line servers considerably. We'll see the DIMM-based XPoint act as a DRAM extension, while the storage bandwidth into that extended memory will be much higher.
In the future data center, servers will run thousands of Docker container instances instead of today's hundreds, leading to a major reduction in server count and space. These new servers will have effectively tens of terabytes of DRAM and all database systems will move to the in-memory model as a result.
Storage is evolving at a phenomenal pace to meet future data center needs. We can expect all-SSD storage products to replace bulk hard drive secondary tier boxes by 2020, while either some form of local PCIe drives or vSAN will host primary storage. Backup and archival storage will move to public clouds, and it's likely that a portion of secondary storage will make the same migration, as vendors such as ClearSky Data solve the latency issues for public cloud storage.
The crucial change in the data center is more a software and philosophic paradigm shift based on these technologies. Software-defined infrastructure will be the vogue in 2020, using the same approaches we already deplore for managing server orchestration. With user control extended to the virtual infrastructure and service on demand, the traditional admin tasks of setup and configuration maintenance are heading to the history books.
By 2020, the data center admin's skill set will comprise of setting up policies and procedures rather than any real IT equipment. Hyper-converged systems will come prepackaged for use, though admittedly at a premium over individual components. This means that they are ready to use and will configure into the infrastructure pool automatically.
It's possible that the explosion of white box units from the original design manufacturers (ODMs) that build in mega volumes for cloud providers may undercut hyper-converged products on price. More likely, the ODMs will offer their own versions of converged and hyper-converged systems. In either case, this will drive down hardware costs and also limit vendor lock-in, nullifying those cozy vendor relationships that can frustrate IT departments.
With future data centers looking so advanced, legacy solutions will face immense pressure. By 2020, we will move away from these antique approaches, creating considerable work in the software admin space during the transition. This won't lead to a huge bonanza for programmers due to substantial pressure to use off-the-shelf solutions. However, these standard software and software as a service alternatives still require a great deal of business process redesign, which should benefit software admins and systems analysts.
At the grander scale, self-orchestrating infrastructure will rapidly lead to users exercising virtual control over private and hybrid cloud deployments, insofar as they will define their virtual infrastructure and software mash-ups. The IT department of the future will need to send staff out to the business departments -- their clients -- to facilitate this and to match the agility that users will expect.
By 2020, we will be in the throes of a full-scale revolution in how information is delivered. Shrinking data centers, growing clouds and full orchestration will change the role of central IT in some fundamental ways and jobs will shift or disappear as a result.
What will the internet infrastructure of the future hold for network operators?
Data center facilities, Ops evolve rapidly
Iceland's shovel-ready sites for future data centers
Purchasing expectations through 2018
The shadow cloud casts on IT jobs