MetaTrader 5 Strategy Tester and MQL5 Cloud Network - page 3

New comment
 
Trolls:

What I don't understand is....

  1. What about the history - will it be the same for all? What if I have downloaded terminal from different brokerage companies with very little history and moreover it has holes in different places?
  2. If the number of instruments is not the same, the example on the server is 12 symbols of the championship. And for testing (multicurrency needs a full matrix of currencies for the indicator to work properly) how in this case ? ....
  3. And thirdly, we already mentioned time, that's why we introduced time UTG - to synchronize somehow ... How will it be like with you? if for example only certain trading hours are tested (for example from 10 to 12 in Moscow) ... The time is different for everyone

1. The entire MQL5 Cloud Network infrastructure is being built for a reason - data caching will be performed both in dispatchers and end-agents. Testing is always performed with reference to a brokerage company and historical data from different companies are never mixed.

2. similar to point 1 - testing is always tied to a brokerage company (trading server), so there are always symbols, they do not overlap with the configuration of other brokerage companies

3. You are confusing the time in the local logs with the testing time. All historical data processes work entirely with server time (charts, trades, symbols, etc.). Given points 1 and 2 (testing is always tied to the trade server) there cannot be any problem with time even theoretically.

 
Interesting:

In your case with your laptop, you should knock out the local cores and run the test on a powerful computer (which is on the local network or whose resources will be as free as possible for the tests).

Thanks, I've roughly figured it out. So, I'm saying that the MQL5 Cloud Network should allow a single test run for such a task - on a remote powerful agent available, preferably on a cluster of supercomputers (those people who can't afford to optimize them should at least have a single test run).
 
-Alexey-:

Thanks for the answers, but a lot is still unclear.

What does it mean: "remote, running in server mode"? I just don't understand - if I install an agent on a second computer using the Metatester component, is this it? What about remote, non-server mode agents - how do I add them?

There are 3 types of agents:

  • local - work on the same computer, run automatically by the terminal when needed (can be used in single tests and optimization)
  • Remote - work in the server mode, waiting for connections from the terminal with authorization by password (can be used in single tests and optimization)
  • MQL5 Network Agents - working through the MQL5 Cloud Network only in optimization mode

This is where we need a supercomputer, or rather a cluster that operates as one core-as one agent, and a network is needed - no one has such a computer at home. Or at least the ability to connect to a powerful machine (as I understand it is possible if I install the agent on a powerful computer, and use it from a laptop during a single run). Exactly for a single run. In fact, it turns out to be the opposite - there is no practical sense in using the MQL5 Cloud Network for massive optimization calculations, if even the initial single run is difficult. Trying out variants is the second case, but a single run is not less important, and even more important for some people.

Any solution has technical validity and economic feasibility.

MQL5 Cloud Network makes sense only for mass optimization calculations.

 
Renat:

There are 3 types of agents:

  • local - work on the same computer, start automatically by the terminal itself when needed (can be used in single tests and optimization)
  • Remote - work in the server mode, waiting for connections from the terminal with authorization by password (can be used in single tests and optimization)
  • MQL5 Network Agents - working through the MQL5 Cloud Network only in optimization mode.

Any solution has its technical validity and economic feasibility.

The MQL5 Cloud Network only makes sense for mass optimization calculations.

Renat, thank you for your clarification. It's clear that this is a specialized solution. I mean that in the open self-regulated network it would be easier to find a powerful agent for the mentioned task, and you're right, you have to look for a remote agent by yourself.
 
Trolls:

What I don't understand is this....

..

If the number of instruments is not the same, the example on the server is 12 symbols of the championship. And for testing (multicurrency needs a full matrix of currencies for the indicator to work properly) how in this case ? ....

..
I can imagine how happy some people will be when they start to get the history for 50 pairs :)
 
Trolls:

What I don't understand is....

  1. What about the history? Is it going to be the same for all? What if the terminal is downloaded from different brokerage companies with very little history, and it has holes in different places?
  2. If the number of instruments is not the same, the example on the server is 12 symbols of the championship. And for testing (multicurrency needs a full matrix of currencies for the indicator to work properly) how in this case ? ....
  3. And thirdly, we already mentioned time, that's why we introduced time UTG - to synchronize somehow ... How will it be like with you? if for example only certain trading hours are tested (for example from 10 to 12 in Moscow) ... The time is different for everyone

1. Every brokerage company has its own history. Either a trader or an Expert Advisor should pay attention to the quality of history when performing it.

2) Testing is performed only for those currencies that are available. Otherwise choose another server (another broker, perhaps only for tests).

You may have an idea about the difference between the terminal's price and the real time you are using it for. So there is no problem, the actual time worked for 60 minutes on 100 cores and the payment will be appropriate.

 
Buter:
I imagine how happy some people will be when they get history of 50 pairs downloaded :)

I can imagine how happy they will be if there are 100-200 pairs and 20-30 of these brokerages (let us take the most popular ones).

Don't even speak about optimization of stock experts (where the number of symbols can be 1000)...

 
All that remains is to run an exchange to determine the cost of processing a standard batch of transactions. For example, set an initial cost of 1 cent per 1000 Tflops (like the cost of one standard lot). Then quote teraflops depending on supply/demand:)))
 
It will not be one huge network where all MT5 users are connected. But it will be split up into chunks by CAs. If you consider that de facto normal history is only available in one place and for a limited number of instruments, it gets a bit sad...
 
You are confused. It will be a huge network working with any broker. You can run tests with any broker's data transparently. Press Start and you are done: all market environment data (symbols, configs, charts) will be downloaded, checked for synchronisation automatically. No need to register brokers in the network, data from different brokers will never cross, the whole server part of the network is a huge data cache, in most cases you will not even have to re-download the history.
12345678910...42
New comment