MetaTrader 5 Strategy Tester and MQL5 Cloud Network - page 28

 
Olegts:

Thanks Renat, I have another question, perhaps somewhere overlapping with Andrei's question, the priority of the metatester.exe process is set to "low", which is basically correct, so as not to interfere with, say, typing in Word, is it possible or do you plan to set the user priority to be low during the day when people work, and full in the evening?

In fact, it is not possible to absolutely control the CPU load when an agent is running.

But lower priority really helps when the user is using the computer for other tasks - in this case less resources are actually allocated to cloud agents (not when running locally!). In other words, you can work on Wordboards without too much trouble.

There is also a schedule table for cloud computing, where you can specify the time of disconnection from the network. For example, during working hours of 08:00-17:00, agents can be disconnected.


Boosting the priority from low to high makes no sense because when there are no higher level tasks, all resources will go to processes with lower priority. This makes it easy for agents to automatically take over 100% of the CPU.

Скачать MetaTrader 5 Strategy Tester Agent для работы в сети MQL5 Cloud Network
Скачать MetaTrader 5 Strategy Tester Agent для работы в сети MQL5 Cloud Network
  • cloud.mql5.com
Подключайтесь к сети распределенных вычислений MQL5 Cloud Network и получайте дополнительный доход круглосуточно — пусть компьютер работает на вас!
 
avoitenko:

Renat, please advise, I have two terminals installed and two managers running. The processor is 4 cores and therefore the agents are 8.

This was done so because the load on the processor with one manager running was insignificant, and with two, it has increased.

Is it possible to do so, and are there no conflicts in the work of managers from different terminals and more importantly a weakness

in terms of maintaining work statistics and possible $ cheating. Thanks in advance.

Are all 8 agents on the cloud network?

In this case it is enough to put only 4 agents. The point is that 8 agents can actually do fewer tasks than 4 agents. Don't forget the fixed costs of each agent (memory, CPU, mass of running threads, etc.).


ps: one skin (computer) can be used to sew 4 full hats (agents), you can sew 8 or even 16, with a corresponding degradation of the result

 
Renat:

Are all 8 agents on the cloud network?

In this case it is enough to put only 4 agents. The point is that 8 agents can actually perform fewer tasks than 4 agents. Don't forget the fixed costs of each agent (memory, CPU, mass of running threads, etc.).


ps: from one skin (computer) you can make 4 full-fledged hats (agents), and you can make 8 or even 16 with the corresponding degradation of the result

It's not for me to confirm the words of a professional, but I know this from practice, it's the purest truth, that's why agents rating was needed.

I used to run 24 agents on a dual core machine, it's true they choke, better one agent per core. Even 2 agents per core is a bit slower.

 
Haven't approached MT5 for a long time. BUT I did rework the owl. Started optimisation and I see that cloud agents are loading history from very distant years (1990 for example). The optimization is being done for the last month. So why are agents downloading so much history?
 
dimeon:
I haven't accessed MT5 for a long time. But i did re-did the owl. Run optimization and see that cloud agents are loading history from very far years (1990 for example). The optimization is being done for the last month. So why are agents downloading so much history?

When starting a test on any time section, the terminal necessarily checks and synchronizes the available history from the server. This is done due to the fact that almost any trader, sooner or later, will request all the available history for testing.

The terminal will give the history to agents in the requested amount, but not the whole available depth. In addition, in 99% of cases, the history of many brokers is already stored on geographically distributed servers of the MQL5 Cloud Network, and from them it is delivered/synchronized to agents.

Remote agents also keep huge caches of historical data for different brokers, which allows them to instantly pass the synchronization stage without downloading history.

We have made a very efficient and cost-effective distributed computing system. To estimate the amount of transferred data on synchronization, look at the logs of agents (not in the terminal, but exactly at the agents in the log files).

 
advise Renat

I can't get an agent to work in the incoming/outgoing traffic column and it stubbornly shows 0kb and doesn't appear in the agents list on the website.
Internet is working through firewall on linux.
You wrote in the very first post that thanks to SSL it can traverse any firewalls.
Maybe I should specify some proxy and authorization settings on it?
or maybe I should open some ports in firewall?

s.e. everything works fine at home
 
leonid100:
advise Renat

I can't get an agent to work in the incoming/outgoing traffic column and it stubbornly shows 0kb and doesn't appear in the agents list on the website.
Internet is working through firewall on linux.
You wrote in the very first post that thanks to SSL it can traverse any firewalls.
Maybe I should specify some proxy and authorisation settings init?
or maybe I should open some ports in firewall?

s.e. everything works fine at home
Try this. It works for me.
 

Hello! Interesting results came out...

 
Karlson:

Hello! Interesting results came out...

Need details.
 

it's all working! Thank you!

Opening, on the Linux firewall, 2000-2001 thresholds for address range 1 to 5.agents.mql5.com helped.