MT4 Strategy Tester memory question | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

MT4 Strategy Tester Question 3 replies

2009.03.11 15:41:15Tester memory handler: tester stopped because not enough memory 19 replies

EA Strategy Tester question 1 reply

Strategy Tester question 1 reply

Strategy Tester Quick Question 1 reply

MT4 Strategy Tester memory question

  • #1
  • First Post: Mar 24, 2011 12:57am Mar 24, 2011 12:57am
If all EA parameters are the same (inputs, timeframe, currency pair) then what does the strategy tester load each time a backtest is started? I am specifically talking about open bar testing.

I made a program that starts up thousands of backtests, one after the other, and I want to know if there is any way to reduce the wait in between tests.
  • #2
  • Mar 24, 2011 2:13am Mar 24, 2011 2:13am
Quoting mfurlend
If all EA parameters are the same (inputs, timeframe, currency pair) then what does the strategy tester load each time a backtest is started? I am specifically talking about open bar testing.

I made a program that starts up thousands of backtests, one after the other, and I want to know if there is any way to reduce the wait in between tests.
In my case, the only way to speed up backtest time was to develop my own backtester taking advantage of modern multicore cpus with parallel processing. On my system, most of the time is spent in loading the tick data from the memory mass storage (approximately 10 seconds for each gigabyte of data on a WD magnetic hard disk). I will soon test if performance improves with an SSD disk.
  • #3
  • Mar 24, 2011 9:47am Mar 24, 2011 9:47am
Your own back tester can't use .ex4 files though.... can it?

I am specifically referring to M5 bars, not tick data.
  • #4
  • Mar 24, 2011 10:07am Mar 24, 2011 10:07am
Quoting mfurlend
Your own back tester can't use .ex4 files though.... can it?

I am specifically referring to M5 bars, not tick data.
My backtester is written in C.
The MQL4 source code of my EA (the only one I work with) can be "almost" directly added and compiled with the backtester sources.
"Almost" means that I have to slightly edit it to allow multithreading execution.
  • #5
  • Mar 24, 2011 2:54pm Mar 24, 2011 2:54pm
Interesting.. Are you parsing the MQL or are you just plugging it in as C?
  • #6
  • Mar 24, 2011 9:35pm Mar 24, 2011 9:35pm
Quoting mfurlend
Interesting.. Are you parsing the MQL or are you just plugging it in as C?
It is compiled as C.
The backtester creates one thread for each simulation point.
The tick data is loaded once and used by all the threads.
  • #7
  • Mar 26, 2011 3:32am Mar 26, 2011 3:32am
I assume that you had to simulate all of the order functions, such as ordersend and ordermodify. How long did it take you, start to finish, to create a basic version of this back tester? How significant was the improvement of execution speed? The application that I am in the process of creating is able to control multiple instances of metatrader, as well as its strategy tester.

It would not take much modification for me to implement multi threaded control over the different instances of metatrader. I am very curious about the difference in execution speed between my method, and your method.
  • #8
  • Last Post: Mar 26, 2011 5:10am Mar 26, 2011 5:10am
Quoting mfurlend
I assume that you had to simulate all of the order functions, such as ordersend and ordermodify. How long did it take you, start to finish, to create a basic version of this back tester? How significant was the improvement of execution speed? The application that I am in the process of creating is able to control multiple instances of metatrader, as well as its strategy tester.

It would not take much modification for me to implement multi threaded control over the different instances of metatrader. I am very curious about the difference in execution...
Of course, I have implemented those functions in my tester.
Up to now not all of them, but only the ones I'm actually using in my EA (therefore I can say it is still in alpha version).
At the moment the source code of the tester is 50kB.
It took me some weeks.
The simulation results are almost 1:1 with the results from MT4 backtester. During this comparison I found some weird behaviour of the MT4 tester so that I'm not going to trust it for serious simulations.
The performance improvement is tremendous, the CPUs are always at 100% load (Intel i7 930) with several threads running simultaneously. I simulate on more or less 10 gigabytes of tick data (you cannot do that with MT4 backtester).
I tried your approach but found big limitations that forced me to think of another solution.
0 traders viewing now