MetaTrader 5 Strategy Tester and MQL5 Cloud Network - page 31

 
notused:

I decided to optimise a simple grid (timer 30 sec, new m1 bar control) on all ticks for two pairs. My 4 i5 cores (PR=160-170) and 8 i7 cores (PR=170-180) optimised for about 90 (!) hours.

...

I am curious. And what period (months, years) is set for optimisation ?
 
tol64:
I would like to ask a question. What period (months, years) is set for optimisation ?
From 01.01.2012 to 26.11.2012 with custom forward from 23.08.2012
 
notused:
from 01.01.2012 to 25.11.2012
Man, it's so long... I want to test some serious developments that need small timeframes on large parts of history (5-10 years). I guess I cannot go anywhere without cloud. ))
 
Have you put agents as services on your local machine too?

On the local machine, there is no need to put agents as services, and it is better to let the tester run them as local agents with normal priority. Agents as services are forced to run at a lower priority, otherwise they consume all resources and the computer cannot be used. If a computer is idle, then the relative lower priority does not matter - all resources are allocated to agents anyway, since there are no other consumers.
 
Renat:
Have you put agents as services on your local machine too?

Agents should not be set as services on the local machine, and the tester should run them as local agents with normal priority. Service agents are forced to run at a lower priority, otherwise they will fully consume all resources and the computer cannot be used. If a computer is idle, the relative lowered priority does not matter - all resources are still given to agents, since there are no other consumers.

I have agents as a service on my local machine, but when I start pimping, I start local agents as well (and I don't disable the service - during my optimization the PR drops anyway and tasks from the cloud don't come). I.e., I don't use my agents-services on this computer. And the priority of local non-service agents is minimal (I will check it again in 2-3 days, after the test is over).

Second machine has service agents - otherwise I can't connect them as remote agents.

But in principle, your position is clear and seems logical. Except maybe add a command line switch to metatester[64].exe to set the default priority for service agents? (for responsible and aware users)

tol64:
Man, this is taking a long time. I want to test some serious developments that need small timeframes on large parts of history (5-10 years). I guess I cannot go anywhere without the cloud. ))
When I was preparing to the Championship, I optimized 12 pairs at a time (using the cloud). And it seems to me that at that time everything ran faster. But then again, "it seems to me" is not a fact :)
 
notused:

...

When I was preparing for the championship, I optimised 12 pairs at a time (using the cloud). And it seems to me that at that time, everything worked faster. But then again, "I think" is not a fact :)
How long was it in one go? And how much did it cost?
 
There are now about 16,000 agents in the cloud and we have worked to make fuller and more efficient use of the clout in sequential settlements. In fact, we have raised the speed of calculations by making fuller use of the network by a factor of 2 to 3.
 
Renat:
There are now around 16,000 agents in the cloud and we have worked to make fuller and more efficient use of the clout in sequential settlement. In fact, the speed of settlements has been raised by a factor of 2 to 3 by making fuller use of the network.
Awesome! Already anticipating how my monster cloud will carry like a feather. Especially when you consider that I won't be ready any time soon and there may be even more agents. ))
 
tol64:
How long is one pass at a time? And how much did it cost?

something like half a day and took 6-8 credits. One pass on a championship iron takes 7-8 minutes.

By the way, local agents do run with the lowest priority on Win7.

I have decided to test and compare how fast 240 passes of my Championship EA (I have only added optimization parameter, it does not affect trading) at different priorities of local agents. I am currently testing the lowest priority. One run takes 4.5 min on the average. I will let you know the results.

 
notused:

I decided to test and compare how fast 240 passes of my Championship EA will run (I only added the parameter for optimization - it doesn't affect the trade) with different priorities of local agents. I am currently testing the lowest priority. One run takes 4.5 min on the average. I will report on the results

On w7 with an unloaded machine, agent priority has no effect on performance (within the margin of error). Not checked on xp64