How it Works a New Computer Cluster like the CapCasc

Let's start with the meaning of the word "CapCasc"

CapCasc - but also CupCask - are the names of a new kind of computer. Let's see how they are composed:

  • CAP is the name given to a "ball", and it stands for Completely Autonomous Processor. They can also be called CUPs (Contained Universal Processor). Or just simply Spheres, Orbs, or Balls.
  • CASC is the name of the remaining parts, and it stands for Communications and Support Carrier. But it can also be simply called Base, or Case.

CapCasc Architecture

In a summary the CapCasc Recipe for a truly Parallel Processing System follows a few basic criterias:
  • True Encapsulation - any given Software Function or set of Functions must encapsulate their own Hardware and Base Software (Firmware, OS, etc.) implementation - as an extension (or completion) of the Encapsulation concept used in Object Oriented Programming Languages. Orthogonally, any given set of Hardware Functions can be implemented within a CAP, but it also has to bring along all its supporting Software
  • Universal Plug-n-Play - the aforementioned True Encapsulation permits the extension of the Plug-n-Play model to all HW and SW objects. In other words, any IT product, either Hardware or Software can be now completely implemented within a CAP - or series of CAPs - and subsequently installed instantaneously and without any adaptation work inside any CapCasc System
  • Parallelization by Replication - True Encapsulation is also the tool of choice, in a CapCasc System, to facilitate Parallelization of existing Algorithms. In fact it makes it extremely easy to replicate any set of Tasks or Functions within any number of CAPs and then distribute the load among them with the usual balancing methods
  • Efficiency by Specialization - finally, True Encapsulation - enabling a straightforward integration of Specialized Hardware and Software within existing Platforms - supports extremely efficient System Designs. For
    Example, an Application can run processing intensive parts of the program, in a Specialized or Reconfigurable Processor, or, a networking SW can work within a Real Time and Security Hardened OS, and both can still operate along with the rest of the system which in turn can use a mainstream OS version and General Purpose CPUs
  • Unlimited Processing Power Scalability - The amount of Processors allowed in a System must be only limited by the available space and Power resources
  • Unlimited Inter-Processor and I/O Bandwidth Scalability - The amount of inter-Processor Data Channels allowed in a System must be only limited by the available space and Power resources
  • Virtually Unlimited Physical Scalability - The system must be able to scale physically to accommodate increasing numbers of Processors and Data Channels
  • Broadcast Data Communication Channels - Along with private peer-to-peer Data Channels, the system must implement an appropriate number of Data Channels, for Broadcasting Purposes
  • Accessible Peer-to-Peer Data Communication Channels - All, or most, of the Data exchanged in non-private peer-to-peer Data Channels, must be readable to all, or most, of the other actors in the System

Additional Architectural and Implementation concepts are:

  • CAPs can be used alternatively for Generic Processing purposes - and they are named Processing CAPs, or to control the system itself - and they are named Control CAPs
  • Control CAPs, implement and control all system level tasks, and arbitrate the access of Processing CAPs to all I/O and Data Communication resources
  • All CAPs have one or more wireless data communication channels
  • Some CAPs can also have one or more wired data communication channels
  • Both Wireless and Wired data channels use RF and Optical signals, which are routed and multiplexed by the CASC Multiplexing Infrastructure
  • The CASC Multiplexing Infrastructure is also used to route all Available Analog and Digital Video Signals of both Processing and control CAPs, so that they can be shared and merged using Overlay and Compositing Techniques
  • All CAPs can have an either wired or wireless power supply - provided by the CASC
  • Control CAPs can selectively switch on and off Processing CAPs
  • All Processing CAPs can both bootstrap and have their firmware updated remotely - from Data Sources managed by the Control CAPs
  • Some Processing CAPs - namely those implementing Proprietary Functions - can implement only a subset of the required features, but still be able to operate within a CapCasc

Features of current CapCasc implementation includes:

  • Processing and Control CAPs, with either AllWinner H3 ARM-Cortex-A7, and H5 ARM-Cortex-A53, or Intel Atom x5-Z8350 Processors both with and 4 cores - based on off-the-shelf SOMs
  • 0.5Gb/s Wireless Optical Data Communication Channels for High-Speed USB and Fast-Ethernet Payloads - implemented with original circuits
  • An original Modular RF and Optical Data Communication Multiplexing and Bridging Infrastructure, supporting 6 CAPs per module - all implemented with custom circuitry, except for the Ethernet and USB switching Hubs which are based on common off-the-shelf modules
  • All Wired Connections - optionally used by a CAP - are consolidated within a single non standard PCI-express connector. A reconfigurable number of USB, Ethernet, Serial, Power and Control connections are supported
  • CapCascOS - a tailored version of Linux, derived from the Ubuntu-Debian Distribution
  • CAPs running Windows
  • Single Server Image, Simulated SMP and Multi-ISA Kernel Aggregation capabilities
  • Compatibility with all the File System Clustering Platforms available in Linux
  • Video Sharing and Compositing through XServer and FreeRDP
  • Hardware Multiplexing of Analog (VGA) and HDMI video signals

Upcoming versions will comprise:

  • CAPs based on off-the-shelf NVidia Tegra-X1 SOMs
  • Integration with Apache Hadoop
  • Open-Hardware Processor Modules based on both Atom and ARM CPUs
  • Optical Wireless Gigabit Ethernet

Here you can download a PDF version of the Product Brief