Public discussion of the formula for calculating the cost of resources in the MQL5 Cloud Network - page 45
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
I see, thanks, and another question, I have 4 cores and judging by the traffic only 1 is active although the rest are on, is it because there are not enough tasks? and also how to enter the context menu?
1. Cores load depends on number of tasks. If only one agent is active and others are idle, it means there are not enough tasks in the cloud.
2. Context menu - same as in Windows - right click on agent list.
Good day to all!
What happened, PR on 8 agnets from 162 to 184 and no yield! For six months 5 quid has barely accumulated! Lately I can say the computer is in deficit. Have people stopped using metatrader 5?
So...
Who's going to say what?
)
Does this "magic system" work ? Does it work or not? From a person who optimizes EAs. It's really cool. Get yourself a PC for a while with 1,500 cores )))
But on the part of the one who offers ? How's it going with the offer, etc. And how often is there work ? I hooked it up last night on interest. Kind of did everything right. But, so far, nothing. )))))))) Nah. I'm not expecting him to earn me at least 30 bucks a month. So far, just for fun, for a week turned on. After that, if nothing. I'll turn it off.)
But. Interesting to know the approximate statistics, who continues to do so. :)
So...
Who's going to say what?
)
Does this "magic system" work ? Does it work or not? From a person who optimizes EAs. It's really cool. Get yourself a PC for a while with 1,500 cores )))
But on the part of the one who offers ? How's it going with the offer, etc. And how often is there work ? I hooked it up last night on interest. Kind of did everything right. But, so far, nothing. )))))))) Nah. I'm not expecting him to earn me at least 30 bucks a month. So far, just for fun, for a week turned on. After that, if nothing. I'll turn it off.)
But. Interesting to know the approximate statistics, who continues to do so. :)
Calculations in the MQL5 Cloud Network
Calculations take into account a tester agent's performance and the time it spends to complete a task. Each testing agent has its own performance index - PR. The higher is CPU productivity, the higher is the index and the more calculations an agent can perform per unit time.
At the moment the accounting of money for the calculations performed is based on the following considerations. The cost of work of a testing agent with PR=100 for an hour is $0.01. One quantum is taken as a unit of work, which is equal to agent's work with PR=1 for 1 ms (1 millisecond). Thus, the cost per quantum is:
The table shows the calculations for running a single-core test agent with PR=100 for 1 hour and for 1 month.
I wonder if the system takes performance slippage into account, depending on the load?
For example 4 core I7 with TurboBoost mode can run at 4GHz without any load (for example) and if load increases up to 100% for each core, then processor frequency drops to 3.4GHz (by 15%) for example. So its performance and PR may be overestimated (if performance test is performed only on 1 core in turn and not on all at once).
And if the processor cooling is bad, then the processor will start reducing the performance even more, so as not to overheat.
As a result, it may turn out that agents from 1-2 core processors (up to I3) without TurboBoost will be more performant or on par with agents with TurboBoost (I5, I7). But I5, I7 processors will be higher in the ranking and will be the first to get the job done slower than I3 agents and at the same time more expensive.
Example
In this situation, I would prefer the I3-6100 agents to the I7-6700I7-6700 Turbo frequency = 4GHz, base frequency = 3.4GHz
I3-6100 always runs at base frequency=3.7GHz
If the rating is measured on each core in turn, I7 will measure at 4GHz and I3 at 3.7.
And at full load, individual agents on I7 will be running at 3.4GHz, i.e. slower than I3, and if it's also more expensive, that would be completely unfair.
I wonder if the system takes performance slippage into account, depending on load?
For example, a 4-core I7 with TurboBoost mode without load may run at 4GHz (for example), but if the load increases to 100% for each core, the processor frequency drops to 3.4GHz (by 15%), for example. And consequently its performance and PR may be overestimated (if performance test was performed only on 1 core in turn and not on all at once).
And if processor cooling is bad, the processor will start to reduce performance even more, so as not to overheat.
As a result, it may turn out that agents from 1-2 core processors (up to I3) without TurboBoost will be more performant or on par with agents with TurboBoost (I5, I7). But the I5, I7 processors will be higher in the rankings and get the job done first, making it slower than the I3 agents and at the same time more expensive.
Example
In this situation, I'd rather have agents on I3-6100 than I7-6700 to do the job.I7-6700 Turbo frequency = 4GHz, base frequency 3.4GHz
I3-6100 always runs at base frequency=3.7GHz
If the rating is measured on each core in turn, I7 will measure at 4GHz and I3 at 3.7.
And at full load, individual agents on the I7 will be running at 3.4GHz, i.e. slower than the I3, and if it's also more expensive, that would be completely unfair.
This is an unreasonably expensive solution for mining in the cloud, such as it should be :) And just to optimize experts... 100 threads and you can optimize anything. Here 32 threads for 300 quid, not bad
I wonder if the system takes into account performance drawdown, depending on load?
For example, a 4-core I7 with TurboBoost mode without load may run at 4GHz (for example), and if the load increases to 100% for each core, the processor clock speed drops to 3.4 GHz (by 15%), for example. So its performance and PR may be overestimated (if performance test was performed only on 1 core in turn and not on all at once).
And if the processor cooling is poor, then the processor will start reducing performance even more to avoid overheating.
As a result, it may turn out that agents from 1-2 core processors (up to I3) without TurboBoost will be more productive or on par with agents with TurboBoost (I5, I7). But at the same time I5, I7 processors will be higher in ranking and will be the first to get the job done slower than I3 agents and at the same time more expensive.
Example
In this situation, I would prefer agents on I3-6100 to I7-6700 to do the job.I7-6700 Turbo frequency = 4GHz, base frequency = 3.4GHz
I3-6100 always runs at base frequency=3.7GHz
If the rating is measured on each core in turn, I7 will measure at 4GHz and I3 at 3.7.
And at full load, individual agents on the I7 will run at 3.4GHz, i.e. slower than the I3, and if it's also more expensive, that would be completely unfair.
Found such an option:
Add to onInit.
int OnInit()
{
if(MQLInfoInteger(MQL_OPTIMIZATION)){
if(TerminalInfoInteger(TERMINAL_CPU_CORES)==1) { return INIT_AGENT_NOT_SUITABLE; }//если 1 то системные процессы будут приоритетнее
if(TerminalInfoInteger(TERMINAL_CPU_CORES)>2) { return INIT_AGENT_NOT_SUITABLE; }//если 4 или 8 то будет TurboBoost. Он при 100% нагр. понизит частоту
if(TerminalInfoInteger(TERMINAL_MEMORY_PHYSICAL)<7000){ return INIT_AGENT_NOT_SUITABLE; }//у современных компов, меньше редко ставят
if(TerminalInfoInteger(TERMINAL_X64)==false) { return INIT_AGENT_NOT_SUITABLE; }//отклонить древние 32 битные компы
}//агент не подходит для проведения тестирования. После возврата этого кода агент больше не будет получать заданий до самого конца данной оптимизации.
.............
return(INIT_SUCCEEDED);
}
Though we need to experiment which option will work faster in practice TERMINAL_CPU_CORES to 2, to 4 or all.
I'd also like to know the frequency of the processor...
Another problem aroused, there is a computer with 8 agents on local network, how can I distinguish agent in local network from agent in cloud? To make sure my computer is 100% used
Really noticed the strong (>10-20%) lag of some agents in the cloud?
With genetics enabled? When the passes (parameter sets) are close in speed?
MQ seems to have struggled with these 'hovering' tasks. It would have been more constructive not to write a patch for yourself, but to reproduce the problem in SR so that it could be solved at cloud level.
Really noticed the strong (>10-20%) lag of some agents in the cloud?
With genetics enabled? When the passes (parameter sets) are close in speed?
MQ seems to have struggled with these 'hovering' tasks. It would be more constructive not to write a patch for yourself, but to reproduce the problem to SD to have it resolved at the cloud level.
I have written to SD, they think the information about what the agent is (local/network/cloud) is important and worthy of inclusion in MT5.
So something has to be worked out ... Apparently we will have to work without screening by number of processors. I.e. just > 1 processor
I have written to SD, they think the information about what the agent is (local/network/cloud) is important and worthy of inclusion in MT5.
So have to figure something out ... Apparently I will have to work without screening by number of processors. I.e. just > 1 processor
Are there really any brakes? I haven't opted with GA in the cloud for a long time.