Comparing optimization speed between MT4 and MT5 - page 3

[Gustavo Correa]

And here it is, the results of optimizing the Moving Average EA over 1, 2, 3 and 4 machines (8, 16, 24 and 32 agents). This time I increased the number of tasks to 1254 by optimizing DecreaseFactor (Start=0,Step=0.3,Stop=3), MovingPeriod (12,1,30) and MovingShift (0,1,5):

Number of machines	Number of Agents	MT4 optimization time	MT5 optimization time	Agents Multiplier	MT4 Performance multiplier	MT5 Performance multiplier
1	8	46:40	86:47	1	1	1
2	16	24:52	44:30	2	1.88	1.95
3	24	16:43	29:51	3	2.79	2.91
4	32	13:43	22:22	4	3.40	3.88

 

I found it very interesting to verify how efficient MT5 is when using several machines/agents. The performance increase multiplier tracks very closely the multiples of agents used, and that is definitely due to the fact that we have one MT5 instance managing all agents, submitting one task per available agent.

The multi-core MT4 setup was not as efficient, mainly because here we had to launch several MT4 instances per agent, with each instance performing a block of tasks (in this case 66 blocks of 19 tasks each). I did test submitting one task per MT4 instance/agent, but the time spent for each task on loading the MT4 instance, performing the calculation and unloading it was counter-productive (as a matter of fact, with 32 agents the optimization time was 14:15)

Anyway, even if the multi-core MT4 optimization was not as efficient as with MT5, that is hardly a problem given how much faster the MT4 setup was compared to MT5.