Using the Cloud for GP Computing - page 2

New comment
 
PCWalker:

Does anyone know about how is it possible to use the MQL Cloud for general purpose computing ? ( Articles )

Like for example using an algorithm on large array divided to multiple agents and returning the results to the source requested.

Thank you for any replies and responses.

PCWalker, I forgot to mention, but last year I proposed a framework to create a Supercomputer using the MT5 Cloud, but in Portuguese forum (see pocket below).

In my opinion, create a standard and generic framework is the way, but I have to slow down in this sense because I was very busy that time.

I was not sure also if this would be a good idea, since we could overload the agents with 'no optimization' tasks.

Anyway, I like too much the idea and count on me to explore more it.

 

Fórum de negociação, sistemas de negociação automatizados e testes de estratégias de negociação

Construindo um Supercomputador usando a Cloud do MT5

figurelli, 2013.12.16 13:56

Que tal criar um projeto totalmente novo, e fora do usual dentro do MT5, mas aproveitando todo o potencial da Cloud Network para construir um Supercomputador genérico?

O potencial da Cloud Network e tendência de crescimento pode ser visto nesse link: https://cloud.mql5.com/en/stats

A ideia seria criar um framework para as mais variadas aplicações de processamento paralelo, e não somente as da área financeira, aproveitando a disseminação internacional da plataforma e a grande quantidade de agentes.

Dessa forma, a proposta é inserir projetos de forecasting e pesquisas dos mais variados tipos.

Não sei até que ponto é possível evoluir nesse sentido, mas minha ideia é criar um case inicial de testes do framework, justamente com um exemplo que tem muito a ver com o nosso idioma.

Nesse caso é possível sonhar com a ideia de construir um Supercomputador usando a Cloud Network do MT5 para projetar o campeão da Copa do Mundo de 2014 ;-)

Quem sabe no futuro não teremos novas aplicações mais genéricas de um Supercomputador rodando na Cloud do MT5.


 
figurelli:

PCWalker, you are welcome.

As far as I know this is not a problem, but a design decision. To explore more agents you must select Slow Complete Algorithm for optimization.

I did select the slow complete option, but still I get only 100-200 agents. Thank you.
 
figurelli:

PCWalker, I forgot to mention, but last year I proposed a framework to create a Supercomputer using the MT5 Cloud, but in Portuguese forum (see pocket below).

In my opinion, create a standard and generic framework is the way, but I have to slow down in this sense because I was very busy that time.

I was not sure also if this would be a good idea, since we could overload the agents with 'no optimization' tasks.

Anyway, I like too much the idea and count on me to explore more it.

 


Yes, I was looking out for that article, do you have the address. I wanted to read it, and see if there are other ideas that I can follow for using the cloud as a Super Computer. Thank you.
 
PCWalker:
Yes, I was looking out for that article, do you have the address. I wanted to read it, and see if there are other ideas that I can follow for using the cloud as a Super Computer. Thank you.
PCWalker, this is just an insight I published here that time, by now the only address I have is my right brain side ;-)
 

For reference, maybe the following chapter "Using the Strategy Tester for Optimization Problems in Mathematical Calculation" in this article https://www.mql5.com/en/articles/239 could be useful.

 
angevoyageur:

For reference, maybe the following chapter "Using the Strategy Tester for Optimization Problems in Mathematical Calculation" in this article https://www.mql5.com/en/articles/239 could be useful.

The title of the chapter is encouraging, anyway the idea is quite simple: the use of 2D/3D graph mode to view the graph of the formula after the optimization, like example below: 

//+------------------------------------------------------------------+
//| Tester function                                                  |
//+------------------------------------------------------------------+
double OnTester()
  {
//---
   double sink=MathSin(x*x+y*y);
//---
   return(sink);
  }

However the challenge of a Supercomputer based on strategy tester regards the integration between each result or step, using the power of parallel processing inside of OnTester().

So, my idea to address this is create a framework, that manages this integration of several parallel results in a way that we can process any parallel algorithm using the same procedures, and solve any problem that a Supercomputer is used to. 

12
New comment