http://www.thegrailsystem.com/forum/ind ... topic=39.0
Since I have posted the results of my benchmark test in September 2009, myself and several Grail users started using MultiCharts in parallel with TS.
I regret to report that intensive usage of the MultiCharts 5.5/6.0 platform has proven that it is unfortunately not as stable as one would have expected.
The instabilities/inconsistencies are of a subtle nature , but when the MC platform is pushed to the limits with software like the GGO/GWFO/CASB, then these problems rear it's head.
Below is some of the problems experienced by Grail users:
- Occasional corrupt records being exported when using the FileAppend & FastFileAppend function (refer
http://www.thegrailsystem.com/forum/ind ... 728#msg728
- As soon as more than 4 cores are being used, then random duplicate records are being exported when using the FileAppend & FastFileAppend function in combination with the BarStatus function.
Refer http://www.thegrailsystem.com/forum/index.php?topic=167 and
http://www.thegrailsystem.com/forum/ind ... opic=253.0
- MC does not support the usage of "Buy/Sell next bar at Open of next bar -/+ X"
in combination with "Buy/Sell next bar at market".
Refer http://www.thegrailsystem.com/forum/ind ... opic=244.0
- Occasional problems with dynamic arrays
Refer http://www.thegrailsystem.com/forum/ind ... opic=223.0
Grail users have independently confirmed that the above problems do not occur if the exact same strategy is run under the same conditions on TS, thus proving that these problems are MultiCharts platform specific.
Most of the problems happen on a random basis which makes it even more problematic.
One simply doesn’t know when you may trust the MC results and when not.
The above discovery is really a great pity because the multi-threading capability of MultiCharts is a great feature.
But, as it stands at this moment, when choosing between MultiCharts and TS, the choice is ultimately between speed and reliability, where MC is the faster platform and TS is the more stable & reliable platform.
When one must make choice between speed and accuracy/reliability, then in my own opinion, accuracy/reliability is most important.
Unfortunately it does not help if a process executes much faster but one cannot rely on the actual figures being produced.
Restricting MultiCharts to using only one core seems to solve most of the above problems, but unfortunately, then most of the performance benefit is also lost.
As explained in my post above, the benchmark showed that on a single core computer, MC is on average only about 7% faster than TS.
Also note that MC5.5/6.0 does not support ELD's created by the latest TS8.8.
Until such stage that MC can handle TS8.8 ELD's, the only way that strategies can be transferred between TS and MC is via text files.
Wouter