Stack it, pack it, virtualize it, and everything in parallel.
As cloud computing takes off for massive computing and business projects the architecture of these ever more specialized services will change quite fundamentally.
I see three trends happening right now which will happen over the next 5 years.
- Hardware is becoming a commodity, by open sourcing designs, from Facebook’s Open Compute, to Blackblaze’s open storage unit. By making the hardware a simple to duplicate, manipulate, and produce commodity, prices are pushed down and economics of scale triggered. This is bad for the margins of the producers, but favorable for the clients and users.
- Processors are getting stacked and the entire architecture of specialized servers is changing in order to accommodate this change. Seamicro which was bought up by AMD, and HP’s Project Moonshot are good examples of how by changing the fundamental architecture of servers around CPU’s one only gains computing power and overall efficiency.
- GPU’s and even FPGA’s are coming into view for supermassive parallel computing. This is still a highly specialized field of computing and most applications need to be rewritten in order to use this computing power. NVIDA’s CUDA project is very promising and is already getting traction.
- And then there is holistic systems thinking and planning of the entire datacenter. People are starting to understand that waste happens whenever secondary output, such as heat, is not used as input for another subsystem of the datacenter. Anything from cooling the hardware in submerged mineral oil cooled stacks to not cooling the racks at all is possible.
And then there is the future after the future, exascale computing.