Public discussion of the formula for calculating the cost of resources in the MQL5 Cloud Network - page 18
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
notused:
All the same, the number of testers and the number of those willing to rent, in my IMHO, will differ by at least 2 orders of magnitude. In these circumstances, no profit or even approximate payback is out of the question, because a good part of the network will be idle or do one run per couple of hours.
1. It is not the number of people, but the number of cores and the number of passes per unit time (say per hour) that should be taken into account.
Even if you consider that there will be two orders of magnitude more cores than those who want to use them, keep in mind that the average number of passes may significantly exceed 10000 (I'd even say in some cases 50000).
I will not even mention the number of symbols or the number of brokers, and people will test more than one or two strategies.
Keep in mind that this is a cyclical process, in which a certain part of the population will optimize their EAs weekly (say, on weekends), and a certain part every day or more than once a week.
2. It should be taken into account that the strategies will be tested by those who do not have MT5 brokerage accounts.
An agent is sent a batch of a certain number of tasks (I once saw in the logs that 21, although I could be wrong) - let it be x. The "seeding" of genetic population is less than a thousand passes (I won't name exact numbers - that's a question for developers) - let it be y. Consequently, y/x agents can be involved. I.e., in my example it's a maximum of 50 agents. Also, speed is determined (roughly) by the slowest agent, since the rest have to wait for it to return a result. Thus using two orders of magnitude more agents is unrealistic in genetic optimization.
As for brute forcing, it's possible to use whole network, but my personal judgement is that Cloud restricts number of agents to be trained.
And also take into account that when network will be paid, there will be less people willing to test compared to free network. For they will stop testing just about anything.
Also, the speed is (roughly) determined by the slowest agent, as the others are waiting for it to return the result. Thus using two orders of magnitude more agents is unrealistic for genetic optimization.
To avoid waiting for slow agents, the cloud server sends the same problem to other guaranteed faster agents when it detects a slow agent.
This solves the problem of bottlenecks.
To avoid waiting for slow agents, the cloud server sends the same task to other guaranteed faster agents when it detects a slow agent.
This solves the problem of slow agents.
And how is an agent determined to be slow?
After all, it could be the case that an agent is just given a slow task.
To avoid waiting for slow agents, the cloud server sends the same task to other guaranteed faster agents when slow agents are detected.
This solves the problem of slowness.
How do you determine if an agent is retarded?
After all, it may be the case that an agent is simply given a retarded task.
The task is not sent to him alone. Probably, by some algorithm of getting average time for a set of agents of this Expert Advisor.
saw my agent being used by someone today... The stats on the website are nil... Why?
How do I use cloud network agents at all? I've tried, but all agents are always in a busy state
To activate them, enable the use of cloud testing agents in the context menu:
Then start the process of optimising your trading strategy.