MetaTrader 5 Platform update build 3660: Improvements and fixes - page 5

New comment
 
fxsaber #:

Try some old builds from the archive (link in profile).


It is important to determine which build the problem started with. Then it will become much easier for developers to find the cause.

Ok, I'll try and report...

Phew b.3082 - b.3661, almost 600 builds - for this hanging test I'll do a "binary search"

 

I tried b. 3223 with the same sad result 31 per hour at the beginning:

Then within the next 2 hours only 22 passes:

And as before some are as fast as a racehorse others as sluggish as a snail.

The longer it runs the slower it gets,  a sorrowful finding.

I'll try now only 5 agents ... :(

 
Carl Schreiber #:

I tried b. 3223 with the same sad result 31 per hour at the beginning:

Then within the next 2 hours only 22 passes:

And as before some are as fast as a racehorse others as sluggish as a snail.

The longer it runs the slower it gets,  a sorrowful finding.

I'll try now only 5 agents ... :(

For example with a Ryzen microprocessor and its AMD Ryzen Master software it is possible to see which threads/cores have priority over the OS.

I find it very useful to choose which threads/cores to select in the tester and not to crash the computer, etc.

amd

 
Miguel Angel Vico Alba #:

For example with a Ryzen microprocessor and its AMD Ryzen Master software it is possible to see which threads/cores have priority over the OS.

I find it very useful to choose which threads/cores to select in the tester and not to crash the computer, etc.


I have Intel and only the Windows Taksmanager which actually looks that way:


while the agents of the tester:


Still build 3223 and again some racehorses and some snails despite the same amount of CPU percentage.

 
Carl Schreiber #:

I have Intel and only the Windows Taksmanager which actually looks that way:


while the agents of the tester:


Still build 3223 and again some racehorses and some snails despite the same amount of CPU percentage.

Is it happening with your EA only or any EAs ?

 
Alain Verleyen #:

Is it happening with your EA only or any EAs ?

I was thinking about this but there was an optimization of this EA with XAUUSD (now it's EURUSD) that ended after two days with 2098 pass (=no. or rows of the xml table) without this problem. This are ~43,7 passes per hour on 10 agents.

 
Carl Schreiber #:

I was thinking about this but there was an optimization of this EA with XAUUSD (now it's EURUSD) that ended after two days with 2098 pass (=no. or rows of the xml table) without this problem. This are ~43,7 passes per hour on 10 agents.

So it doesn't always happen with your EA.

But that doesn't really answer the question

So what about an other EA ?

What time it takes for a simple backtest ?

 
Alain Verleyen #:

So it doesn't always happen with your EA.

But that doesn't really answer the question

So what about an other EA ?

What time it takes for a simple backtest ?

Well I want this EA to be optimized. It is currently running since 2 weeks without any problem in a version that paints the theoretical trades on the charts and it looks quite good.

That's why I first try older builds ..

 
Carl Schreiber #:

Well I want this EA to be optimized. It is currently running since 2 weeks without any problem in a version that paints the theoretical trades on the charts and it looks quite good.

That's why I first try older builds ..

Are you running the genetic algorithm or a complete test ? Try with a range of inputs with a complete test , measure speed , then pick another range way different and measure speed again.

Also do you have Cos / Sin / Arctan2 calcs in the ea ? The only time i've noticed a gradual slowdown during the test was with these math functions  

Also , if the first passes are faster , take one set of inputs from an early pass , then take a set of inputs from a "slow" pass and run them independently . Measure speeds 
 

I found the problem, ash on my head :(

  1. I enabled full Optimization log in the Tester's Journal and waited ..
  2. After half a day I found an entry: returned result of .... in 6:..:..
  3. I stopped the optimization and run this set up in visual mode to get time and date of the temporal black hole
  4. started the debugger close to that black hole
  5. and found that I just had to replace a = by >= which in 1 out 100 cases fills a buffer again and again and held the loop there longer and longer ...
12345678910
New comment
Reason: