ForumSections

Discussion of article "MQL5 vs QLUA - Why trading operations in MQL5 are up to 28 times faster?" - page 4

New comment

If we rephrase the problem as "how often MQL5 code is called from the terminal kernel", then everything should be measured wrong:

  1. We should remove all unnecessary MQL5 code and leave a couple of counters in one bare function to minimise the cost of the measurement itself
  2. Find a suitable OnXXXX entry point and a method of forced generation of calls from the kernel. Here we need to make a separate test build for metering.

In this case, having built a polygon, you can really measure all the overheads on MQL5 calls and get an interesting characteristic. As is our custom, then optimise everything several times.

This is an interesting task and we will deal with it.

fxsaber:


I got it!

Yes, we do.

Sorry, I overlooked it.

Renat Fatkhullin:

If we rephrase the problem as "how often MQL5 code is called from the terminal kernel", then everything should be measured wrong:

  1. We should remove all unnecessary MQL5 code and leave a couple of counters in one bare function to minimise the cost of the measurement itself
  2. Find a suitable OnXXXX entry point and a method of forced generation of calls from the kernel. Here we need to make a separate test build for metering.

In this case, having built a polygon, you can really measure all the overheads on MQL5 calls and get an interesting characteristic. As is our custom, then optimise everything several times.

This is an interesting task and we will do it.

Thank you for not leaving such geeky things without attention!

Probably, we need a separate build for measuring the speed of arrival of quote packets. Just what I wanted to measure.

[Deleted]
Renat Fatkhullin, fxsaber - thanks for the clarification.
fxsaber:

Please do another MT5 speed test on a real account

Forum on trading, automated trading systems and testing trading strategies

Testing 'CopyTicks'

fxsaber, 2016.09.13 11:11 AM

if through OrderSendAsync send two limiters (BuyLimit1_price < BuyLimit2_price) inside the spread, will the exchange generate two consecutive ticks with bid-price improvement at the same time (with 1ms accuracy)?

Let me explain what this will do.

When BuyLimit1 is sent inside the spread, a tick will come to us from the exchange with the time of its birth (it will generate a new Bid). After BuyLimit2 - another tick with its birth time. The difference of these two times is the speed characteristic of MT5 trading orders delivery to the exchange.

In order to make minimal monetary losses from such an experiment, you can send not two BuyLimits asynchronously, but BuyLimit and SellLimit inside the spread and choose some poorly liquid trading instrument.

Before comparing, we need to "finish" the whole complex of trading operations

(receiving data, justification of transactions, receiving confirmation of transactions).

Added Build 1395, real

2016.09.14 17:30:03.100 Trades  'xxxxx': sell limit 5.00 BR-12.16 at 47.56
2016.09.14 17:30:06.849 Trades  'xxxxx': accepted sell limit 5.00 BR-12.16 at 47.56
2016.09.14 17:30:06.849 Trades  'xxxxx': sell limit 5.00 BR-12.16 at 47.56 placed for execution in 3750.466 ms
2016.09.14 17:30:06.851 Trades  'xxxxx': cancel order #44309414 sell limit 5.00 BR-12.16 at 47.56
2016.09.14 17:30:10.014 Trades  'xxxxx': deal #29388895 sell 1.00 BR-12.16 at 47.56 done (based on order #44309414)
2016.09.14 17:30:10.014 Trades  'xxxxx': exchange buy 1.00 BR-11.16 at market
2016.09.14 17:30:10.385 Trades  'xxxxx': deal #29388914 sell 4.00 BR-12.16 at 47.56 done (based on order #44309414)
2016.09.14 17:30:12.374 Trades  'xxxxx': accepted exchange buy 1.00 BR-11.16 at market
2016.09.14 17:30:12.375 Trades  'xxxxx': exchange buy 1.00 BR-11.16 at market placed for execution in 2360.902 ms
2016.09.14 17:30:12.398 Trades  'xxxxx': deal #29389024 buy 1.00 BR-11.16 at 47.25 done (based on order #44309578)
2016.09.14 17:30:12.401 Trades  'xxxxx': exchange buy 4.00 BR-11.16 at market
2016.09.14 17:30:13.006 Trades  'xxxxx': accepted exchange buy 4.00 BR-11.16 at market
2016.09.14 17:30:13.007 Trades  'xxxxx': exchange buy 4.00 BR-11.16 at market placed for execution in 606.852 ms
2016.09.14 17:30:13.009 Trades  'xxxxx': deal #29389140 buy 4.00 BR-11.16 at 47.25 done (based on order #44309644)
2016.09.14 17:30:13.015 Trades  'xxxxx': buy limit 5.00 BR-12.16 at 47.74
2016.09.14 17:30:13.357 Trades  'xxxxx': accepted buy limit 5.00 BR-12.16 at 47.74
2016.09.14 17:30:13.357 Trades  'xxxxx': buy limit 5.00 BR-12.16 at 47.74 placed for execution in 342.736 ms
2016.09.14 17:30:13.668 Trades  'xxxxx': buy limit 5.00 BR-12.16 at 47.79
2016.09.14 17:30:13.712 Trades  'xxxxx': accepted buy limit 5.00 BR-12.16 at 47.79
2016.09.14 17:30:13.716 Trades  'xxxxx': buy limit 5.00 BR-12.16 at 47.79 placed for execution in 48.228 ms
2016.09.14 17:30:13.718 Trades  'xxxxx': deal #29389165 buy 5.00 BR-12.16 at 47.79 done (based on order #44309680)
2016.09.14 17:30:13.721 Trades  'xxxxx': exchange sell 5.00 BR-11.16 at market
2016.09.14 17:30:13.740 Trades  'xxxxx': accepted exchange sell 5.00 BR-11.16 at market
2016.09.14 17:30:13.741 Trades  'xxxxx': exchange sell 5.00 BR-11.16 at market placed for execution in 20.867 ms
2016.09.14 17:30:13.778 Trades  'xxxxx': deal #29389166 sell 5.00 BR-11.16 at 47.29 done (based on order #44309682)

Forum on trading, automated trading systems and testing of trading strategies

FORTS. Questions on execution

Renat Fatkhullin, 2016.08.23 16:35

You just need to wait 5-10 ms and try again.

The thing is that you get transaction confirmation immediately, but full transaction details arrive asynchronously after that. It can take from 0 to N ms, usually within 1-2 ms (depends on ping of course).

Because of this circumstance, the comparison with QLUA is not in favour of MT5.
prostotrader:

Before comparing, we need to "finish" the whole complex of trading operations

(receiving data, justification of transactions, receiving confirmation of transactions).

Added Build 1395, real


Just did a similar operation manually: 11.7 ms

2016.09.14 18:17:26.298	Trades	'10321': order #44324961 buy limit 1.00 / 1.00 BR-12.16 at 47.00 done in 11.759 ms
2016.09.14 18:17:26.295	Trades	'10321': buy limit 1.00 BR-12.16 at 47.00 placed for execution
2016.09.14 18:17:26.293	Trades	'10321': accepted buy limit 1.00 BR-12.16 at 47.00
2016.09.14 18:17:26.286	Trades	'10321': buy limit 1.00 BR-12.16 at 47.00
Renat Fatkhullin:

You do not want to see the real report, but use your own glitchy method of measuring transaction time.

And you do not want to admit your mistakes, preferring to believe monstrous figures from your erroneous script.

I have just made a similar operation manually: 11.7 ms.

What are you talking about, Renate?

This is an excerpt from Temninal's log, not my own log!

fxsaber:
Because of this circumstance, it is necessary to slightly adjust the comparison with QLUA not in favour of MT5.

There is no need to adjust anything. My phrase was for the general case "in your internet and your ping can be anything and in reality depending on your network you will get a transaction in 0-N ms". And it's more likely to be 0 ms than more.

Here is the verification code with control of all transactions in asynchronous mode:

ulong ExtTicks=0;
//+------------------------------------------------------------------+
//| Expert initialisation function|
//+------------------------------------------------------------------+
int OnInit()
  {
//--- output the ping to the trade server in milliseconds
   PrintFormat("AsyncTradesTest: last ping %.2f ms, build %d",TerminalInfoInteger(TERMINAL_PING_LAST)/1000.0,TerminalInfoInteger(TERMINAL_BUILD));
//---
   return(INIT_SUCCEEDED);
  }
//+------------------------------------------------------------------+
//||
//+------------------------------------------------------------------+
void OnTradeTransaction(const MqlTradeTransaction &trans,
                        const MqlTradeRequest &request,
                        const MqlTradeResult &result)
  {
   Print(EnumToString(trans.type)," ",trans.symbol,": ",trans.price," ",result.order," time: ",(GetMicrosecondCount()-ExtTicks)/1000.0," ms");
  }
//+------------------------------------------------------------------+
//| Expert tick function|
//+------------------------------------------------------------------+
void OnTick()
  {
   static bool started=false;
//--- run the series once on the first tick
   if(!started)
     {
      started=true;
      //--- fill in the enquiry
      MqlTradeRequest req={};
      MqlTradeResult  res={};

      req.volume      =1;
      req.symbol      =_Symbol;
      req.price       =SymbolInfoDouble(_Symbol,SYMBOL_ASK);
      req.sl          =0;
      req.tp          =0;
      req.deviation   =100;
      req.type_filling=ORDER_FILLING_RETURN;
      req.action      =TRADE_ACTION_DEAL;
      req.type        =ORDER_TYPE_BUY;
      req.magic       =2016;
      //--- run the operation loop
      ExtTicks=GetMicrosecondCount();
      Print("Start...");
      OrderSendAsync(req,res);
     }
  }

Here is its output on a real account just now:

2016.09.14 18:32:53.234 Test (BR-12.16,H1)      TRADE_TRANSACTION_HISTORY_ADD BR-12.16: 0.0 0 time: 11.455 ms
2016.09.14 18:32:53.232 Test (BR-12.16,H1)      TRADE_TRANSACTION_DEAL_ADD BR-12.16: 47.13 0 time: 8.778 ms
2016.09.14 18:32:53.231 Test (BR-12.16,H1)      TRADE_TRANSACTION_ORDER_DELETE BR-12.16: 0.0 0 time: 8.457000000000001 ms
2016.09.14 18:32:53.231 Test (BR-12.16,H1)      TRADE_TRANSACTION_ORDER_UPDATE BR-12.16: 0.0 0 time: 8.202999999999999 ms
2016.09.14 18:32:53.231 Test (BR-12.16,H1)      TRADE_TRANSACTION_REQUEST : 0.0 44326441 time: 8.151 ms
2016.09.14 18:32:53.228 Test (BR-12.16,H1)      TRADE_TRANSACTION_ORDER_UPDATE BR-12.16: 0.0 0 time: 5.196 ms
2016.09.14 18:32:53.228 Test (BR-12.16,H1)      TRADE_TRANSACTION_REQUEST : 0.0 44326441 time: 5.171 ms
2016.09.14 18:32:53.227 Test (BR-12.16,H1)      TRADE_TRANSACTION_ORDER_ADD BR-12.16: 0.0 0 time: 4.184 ms
2016.09.14 18:32:53.223 Test (BR-12.16,H1)      Start...
2016.09.14 18:32:49.753 Test (BR-12.16,H1)      AsyncTradesTest: last ping 2.15 ms, build 1417

You should read it from bottom to top.

It shows all stages of transactions with the cumulative total of time spent from the start. By transaction types you can see what was added to the terminal and when.

The total time is 11.45 ms.

1234567
New comment