I think that a real clustering solution may reach home market and enterprise with a unique solution with the consequence of a wide standards and compatibility.
I post here with the hope that someone else already thought about this or if not someone will.
As project ara aim to, I'm try to understand if and how is theoretically possible to assemble a set of hardware that allow end users to simply assemble their hardware at the size he care. (18 cpu for real time application instead of 2 TB of ram for managing in real time enormous file; free your immagination)
Based on the block structure as: cpu, ram, rom, I/O peripherals i ended with the solution of a connector that itself handle the route process of the signal in a 4 way node for all of the mentioned blocks (cpu, ram, rom, I/O peripherals).
My low knoledge do not allow me to practically evaluate a similar connector and i was not able to find out something similar on the market (beside the fact that a lot of routing device are sold like parallel, usb or ethernet hub/router).
So here is the question: is it realistic to think about a connector (4 way) that include a bridge chip that rute the signal in the right place and provide a universal bus system that self recognizes the expansion ?
Writing this post i'm thinking in an open sourced solution that provide a similar service as the unipro (project-ara) is providing for the future modules.
If i don't get wrong inside the epiphany core there's a piece that route the data to the right core, is it possible to include that piece in a 4 way connector for example, or in a sort of backplane (more than 4 way) where the expansion device are all different as function but the connector still the same ?
In this way we could have a good possibility to scale, but it's to far for me to evaluate the problem on designing something like this.
The pourpuse of this is to scale from a simple device (raspi, arduino, old gsm phones, basic android tablet) to a big and complex system as the computer that handle the weight of internet (mainframes, structures that allow cloud services, ecc) as simple as a kid build stuffs with bricks
.The bottle neck at my eyes is the "bus" system, mean the structure that recognizes and allow component to communicate eachother, so a parallel and universal structure (as i understand that unipro do) cold be the first step to a real scalable system.
So, just for the try to explain myself better try to imagine a small board as parallella , surrounded by 4 universal connector that allow you to expand the device with other same boards, or just connect the the end termination needed (video, audio, general peripherals or what the user need).
This may result in a lot more component but in a lot more versatility. the fact that the connector is unique and universal should allow to a mass production the minimization of the cost for it.
Lot of simple devices interconnected at my eyes are better than a little single and complex device that do everything; the single device may give more performance but surely if it is too complex he is going to end is life late or soon, simple device can have a longest term of support.
for example : what should be done to make the provided samtec connector on parallella working as the unipro connection system on ara-project ?
Thanks for your patience, your considerations and your hints in addition to the thanks for your great job you did and still doing.
