Reflections on Foggy Computing
Good day.
On Habré already flashed an article on this topic. I want to try to develop this theme, and share my thoughts - how could it be.
So, what is “fog computing”, or “fog computing”. These are calculations based on a distributed infrastructure with non-guaranteed availability. Topologically, this is a mesh (mesh) network with dynamic routing, the nodes of which are relatively homogeneous computers in computing power.
Ideally, in the era of “foggy computing,” computer nodes are literally everywhere - under your feet, in the air, on the street ... They are so tiny and cheap that they can be carried with them in kilograms. In our time, this will most likely be some kind of software environment that consolidates the resources of a multitude of virtualized “drops” and allows programs written for cross-platform environments to run on such a parallel machine — platform dependency in such an environment would be deadly. Most likely, we will talk about Java, CLR, Python, JavaScript ...
As for the hardware base - it is unlikely that these will be pure microcontrollers - the resources of the microcontroller are very limited and insufficient for creating a shared environment. But most likely, the “drops” will be served in parallel by several consumers. But the very concept of execution of such devices is close to the concept of a microcontroller, or SoC. In my mind, the chip of such a device would look like this: a multi-core CPU, GPU, 3-4 GB DRAM, and 32 GB SSD plus minimum peripheral logic are placed on the same chip. Out comes out a minimum of interfaces - SATA and SD for “programming” (recording OS on SSD), HDMI + VGA + USB in case the system is used as a workstation, LightPeak or 10GE for intersystem communications. If it is a control node, then GPIO. Perhaps - some local radio interface. In general, such a system seeks to design Atom CE or Raspberry Pi - a minimalist energy-efficient computing system. Of course, there is a place for the battery and for the wireless power receiver.
In the first place, the software foundation is likely to include a hypervisor capable of consolidating the power of the hazy network and presenting them as a single multiprocessor system. This approach was once used in OS OpenMosix. The resources requested by the consumer (another “drop”) will be allocated in the form of a virtual machine, in which, for example, the bare-metal Java machine is launched. It deploys the application package, the application starts, and so on ...
It is particularly worth mentioning the role of drop computers as a “user terminal” - this is the only scenario when some kind of peripherals are connected to the “drop”. I see this as some kind of device shell, like Motorola's SmartDocks: a shell-smartphone, a shell-tablet, a shell-laptop, a shell-desktop. The laptop shell and the desktop shell may well accommodate several modules, and thus carry a multi-node private cluster, a “cloud of fog”. And the person turns into a real PAN.
')
With this approach, the system upgrade turns into a simple connection of new nodes. If the node-drops will be the size of a microSD card, the future system unit may well be a three-liter jar with plates of wireless power at the ends, and a thin thread of optics to a huge monitor.
In such a situation, “three kilograms of counterfeit servers” no longer look like nonsense.
Summarizing the above, it can be noted that we cannot see the “hazy PCs” and the “hazy networks” until the following requirements are met:
- Availability of highly efficient (in W / MIPS) SoC common architecture (ARM, x86, AMD64) with minimal strapping at an extremely low price ($ 2 - $ 10)
- Standardization and distribution of consolidating hypervisors
- The widespread adoption of IPv6 and high-speed interfaces - both cable and wireless.
- Availability of cheap and capacious sources of autonomous power supply (at least an order of magnitude more capacious than lithium-ion batteries)
The issues of authentication, authorization, and sharing of the “drop” resources by several “foggy networks” are still behind the scenes.
On Habré already flashed an article on this topic. I want to try to develop this theme, and share my thoughts - how could it be.
So, what is “fog computing”, or “fog computing”. These are calculations based on a distributed infrastructure with non-guaranteed availability. Topologically, this is a mesh (mesh) network with dynamic routing, the nodes of which are relatively homogeneous computers in computing power.
Ideally, in the era of “foggy computing,” computer nodes are literally everywhere - under your feet, in the air, on the street ... They are so tiny and cheap that they can be carried with them in kilograms. In our time, this will most likely be some kind of software environment that consolidates the resources of a multitude of virtualized “drops” and allows programs written for cross-platform environments to run on such a parallel machine — platform dependency in such an environment would be deadly. Most likely, we will talk about Java, CLR, Python, JavaScript ...
As for the hardware base - it is unlikely that these will be pure microcontrollers - the resources of the microcontroller are very limited and insufficient for creating a shared environment. But most likely, the “drops” will be served in parallel by several consumers. But the very concept of execution of such devices is close to the concept of a microcontroller, or SoC. In my mind, the chip of such a device would look like this: a multi-core CPU, GPU, 3-4 GB DRAM, and 32 GB SSD plus minimum peripheral logic are placed on the same chip. Out comes out a minimum of interfaces - SATA and SD for “programming” (recording OS on SSD), HDMI + VGA + USB in case the system is used as a workstation, LightPeak or 10GE for intersystem communications. If it is a control node, then GPIO. Perhaps - some local radio interface. In general, such a system seeks to design Atom CE or Raspberry Pi - a minimalist energy-efficient computing system. Of course, there is a place for the battery and for the wireless power receiver.
In the first place, the software foundation is likely to include a hypervisor capable of consolidating the power of the hazy network and presenting them as a single multiprocessor system. This approach was once used in OS OpenMosix. The resources requested by the consumer (another “drop”) will be allocated in the form of a virtual machine, in which, for example, the bare-metal Java machine is launched. It deploys the application package, the application starts, and so on ...
It is particularly worth mentioning the role of drop computers as a “user terminal” - this is the only scenario when some kind of peripherals are connected to the “drop”. I see this as some kind of device shell, like Motorola's SmartDocks: a shell-smartphone, a shell-tablet, a shell-laptop, a shell-desktop. The laptop shell and the desktop shell may well accommodate several modules, and thus carry a multi-node private cluster, a “cloud of fog”. And the person turns into a real PAN.
')
With this approach, the system upgrade turns into a simple connection of new nodes. If the node-drops will be the size of a microSD card, the future system unit may well be a three-liter jar with plates of wireless power at the ends, and a thin thread of optics to a huge monitor.
In such a situation, “three kilograms of counterfeit servers” no longer look like nonsense.
Summarizing the above, it can be noted that we cannot see the “hazy PCs” and the “hazy networks” until the following requirements are met:
- Availability of highly efficient (in W / MIPS) SoC common architecture (ARM, x86, AMD64) with minimal strapping at an extremely low price ($ 2 - $ 10)
- Standardization and distribution of consolidating hypervisors
- The widespread adoption of IPv6 and high-speed interfaces - both cable and wireless.
- Availability of cheap and capacious sources of autonomous power supply (at least an order of magnitude more capacious than lithium-ion batteries)
The issues of authentication, authorization, and sharing of the “drop” resources by several “foggy networks” are still behind the scenes.
Source: https://habr.com/ru/post/136536/
All Articles