We have constantly been asked for more technical information about Prometheus59 implementation, as well as its difference from already existed solutions, and today we are happy to release an update with some technical insights.
Simplicity is the Best Marketing Tool
Since Prometheus59 is based on distributed computing, one of the main milestones is the simplicity of the task distribution process for users. Simplicity means that users without any technical background can easily deal with Prometheus59 and have no problems with renting and providing computing power.
After a line of tests completed with a part of the alpha-code, we have revealed certain possible implementation-stage issues with the concept of task distribution when dealing with heterogeneous tasks.
We have consulted with third-party developers, as well as made a private survey among potential users, to gather the feedback about the concept itself. The main requirements and demands were directed at the lack of simplicity for end users. Based on the feedback, we decided that, although the concept of "compilation to an own execution language from outer languages" looks attractive and universal, it can be an obstacle for the mass use of Prometheus59at the initial stage.
Our initial concept could have provided significant benefits in the future, but at the initial stage it would require a serious change in the code of the tasks that are currently being solved on existing supercomputers.
Simply put, users would have had certain knowledge, and that’s a great barrier. The likelihood that large customers will make such changes is nothigh, so we decided to develop a more universal concept.
MPI Standard Supported
For the successful development of the project, it is necessary to have a working model with clients at the beginning. To make it possible, users should be able to run their tasks without changes or at least without major changes to the code, and without having to learn the features of coding for a particular system. At the same time, the existing code intended for use on supercomputers was written using standardized MPI or PVM protocols.
Proceeding from all available options, we decided to introduce"computing nodes ranking" in order to adapt the system to one of the standard protocols – MPI, which will allow one to directly process tasks on Prometheus59 the same way they are processed now on expensive supercomputers.
This mechanism is in active development. It will allow:
1) Implementation of the most efficient task distribution protocol between the computers of Prometheus59. The "most efficient task distribution" means that all types of computers, from old to modern, can support distributed calculations. The basis of this protocol has already been invented and is being actively developed and tested (we cannot give you more updates on the protocol for information security reasons).
2) Adaptation of our system to universal software architecture. This will solve most problems, including those sharpened for different processor architectures.
3) Using already available GPL-licensed tools for creating cluster supercomputers.
All together it will
a) Speed up Prometheus59 launch.
b) Provide more convenience for customers, as the main interfaces and principles will be similar to traditional clusters, and customers won’t need to make significant changes in task shaping.
Follow us to stay updated: https://discordapp.com/invite/Ac8cHV5