Interesting topic for many: what's new in MetaTrader 4 and MQL4 - big changes on the way - page 69

[TheXpert]

C-4:

I.e. it is true that the time to deploy a lightweight structure consisting of a single long value has decreased manifold, from 9 seconds to 2.35, but the time of the execution itself remains almost the same

And if we divide it by 8 (cores)?

[Vasiliy Sokolov]

TheXpert:
What if you divide by 8 (cores)?

If you divide it, it is possible to reach speed of 100 000 000 bars, but still, we are reasoning too abstractly and speculatively. In a real tester, there will anyway be costs on function calls, deploying and reading data from the disk, mathematical calculations inside strategies, sending signals to buy and sell, recording and storing of intermediate data and much, much more. A single call of the TradeSystem() function in the first example increases strategy execution time by 40%. So even if we divide the result by 8, we will only get closer to the declared limit, but it was stated that it was reached within 5 hours without any help from a programmer. And we won't know for a long time yet if hrenfx meant the performance in multithreading or single-threaded mode.

[Vladimir Gomonov]

TheXpert:
And if you divide it by 8 (cores)?

No, hrenfix has a single-threaded tester, it is written in his post.

C-4:

Some of the hrenfx codes did come across to me - the code is very high quality, I can't say anything. I still use some of them. But don't mix flies with cutlets. You, like gerica, are writing without even understanding the test I'm offering. Either from lack of deep C knowledge, or for some other reason you are emphasizing that one bar for hrenfx is only two integer longs. In reality, we pass a pointer to the structure describing the bar; the structure itself isn't passed by value, which means that turning on and off the number of items in the bar won't affect performance in any way. Note, I'm talking about performance time of the stratum itself, I'm ignoring time of array filling.

Here's the performance result if you leave only one value in the structure itself:

I.e. indeed, the time to deploy a lightweight structure, consisting of a single long value decreased many times, from 9 seconds to 2.35, but the execution time itself remained almost the same (even slightly increased, because I started to call rand() in the if check). If the engine delegates execution to the strategist, which actually happens, the execution time becomes even longer and the size of the structure describing the bar has absolutely nothing to do with it.

So, if you're going to say something again, study C for starters and then we'll have something to talk about.

I don't want to discuss the discrepancy found less than twice, // which can be written off, for example, to differences in compilers and processors.

Ivan's performance is close to real one, for light strategies, and it's quite motivating for writing simple "calculators" for each of his strategies.

For that productivity is huge compared to a universal mashup of regular tester. That's exactly what Ivan wanted to say, and not to Renat, but to you, me and other "users" who are waiting "under the sea of weather".

[михаил потапыч]

C-4:
not a programmer, in just five hours.

I do not remember that he said he was not a programmer, but he has long been known here as a programmer among other things.

[TheXpert]

MetaDriver:
No, hrenfix has a single-threaded tester, it's written in his post.

He also has the same instructions on how to unpaired it.

[TheXpert]

Mischek:
I do not remember that he said he was not a programmer, but he has long been known here as a programmer among others.

He is not a pro, and unlike some people who think they are pros, he says so, although at a higher level than most of the top jobs.

[Vladimir Gomonov]

C-4:

1) ...... But it was also stated that it was achieved, as it were, with half a bang, by a non-programmer, in just 5 hours.

2) And we will not know for a long time whether hrenfx meant performance in multithreaded or single-threaded mode.

1) The result was achieved in five hours by a man who writes the tester not for the first time, i.e. a very experienced "non-programmer". For example,one of his feats, three years old already.

2) Read the original post more carefully and you'll find out right away. "the crank is not a reader..." ? ;)

[михаил потапыч]

Crenfix's witnesses thread ))

[михаил потапыч]

TheXpert:
He is not a pro, and unlike some who think they are pros, he says so, even though he is probably above most of the top jobs in terms of level.

Top joba sucks.)

[Vasiliy Sokolov]

MetaDriver:
No, hrenfix has a single-threaded tester, it's spelled out in his post.

Yes, he wrote about that. The other thing that is not clear is what he meant by the performance of 100,000,000 bars. Is that the combined performance of 8 testers or the performance of 1 tester running on a single CPU core.

With all due respect to hrenfx, MetaQuotes is one of the fastest (if not the fastest) MetaTrader4 tester in terms of optimization, so Renat knows what he's talking about and would never call hrenfx a fibber and a liar (or whatever) just because hrenfx has a different opinion about what a trader needs. A simple cycle run showed that the claimed speeds are at the limit, at Renat, the fastest MT4 tester shows results an order of magnitude lower. So what is the point or the argument? It is clear that even if the claimed result is true, then we are not talking about any full-fledged and universal tester, but we are talking about a cycle for, proudly called "tester" at best.