Possible performance bug in Portfolio Backtester / Optimizer

Questions about MultiCharts and user contributed studies.
SiliconTiger
Posts: 24
Joined: 21 Feb 2009

Possible performance bug in Portfolio Backtester / Optimizer

Postby SiliconTiger » 16 Jul 2009

I am running MC Version 5.0 Release (Build 2369) on Windows XP Pro SP3 on a quad-core Intel CPU. I noticed when using the genetic optimizer that for awhile at the beginning of an optimization run that all 4 CPUs are max'ed out during the run (looking at the Performance tab of the Windows task manager), but after several minutes only two of the CPUs are being used. In a specific run, the genetic optimizer starts out estimating the total run length will be 20 minutes, and it runs for ~5 minutes or so using all 4 CPU cores. But then after about 5 minutes only 2 CPU cores are being utilized and the optimizer's estimated total run length bumps up to ~28-29 minutes. Once it reaches this state of only using 2 CPU cores it remains that way until the optimizer run completes.

Is this expected behavior? Seems wrong not to be able to utilize all available CPU resources through the entire length of the optimizer run.

Thanks for any feedback on this issue.

geektrader
Posts: 100
Joined: 17 Jul 2009
Location: Germany

Postby geektrader » 17 Jul 2009

I can absolutely confirm this. I am having a Quad Core CPU as well and for about the first 10 to 20% of Genetic Optimization all 4 Core´s are being used, then only 1 core and Estimated Time Left from this point does about quadrable. I´ve already reported this a while ago as a bug but never got an answer unfortunately:(

Hope this gets fixed.

User avatar
TJ
Posts: 6551
Joined: 29 Aug 2006
Location: Global Citizen
Has thanked: 966 times
Been thanked: 1892 times

Postby TJ » 18 Jul 2009

is that a home-built computer?
did you OC it?

geektrader
Posts: 100
Joined: 17 Jul 2009
Location: Germany

Postby geektrader » 18 Jul 2009

Attached are 2 screenshots where you can see that for the first few percent of genetic optimization all 4 Core´s are being used. Then after some percent, it drops down to only 25% load and estimated time left increases.

Interesting is: this doesn´t happen with Exhausive Search! There all 4 Core´s are fully used all the time.
Attachments
opimizerproblem.GIF
opimizerproblem.GIF (100.51 KiB) Viewed 244 times
optimizerproblem2.GIF
optimizerproblem2.GIF (100.41 KiB) Viewed 243 times

geektrader
Posts: 100
Joined: 17 Jul 2009
Location: Germany

Postby geektrader » 18 Jul 2009

No it´s not OCed and the computer is self-built. However, other applications, as well as the "Exhausive Search" from Multicharts do utilize all 4 Core´s permanently without problems. Only the Genetic Optimizer within Multicharts creates this problem, and not only for me as you see.

SiliconTiger
Posts: 24
Joined: 21 Feb 2009

Postby SiliconTiger » 18 Jul 2009

My computer is completely off-the-shelf -- Dell with Intel Core™2 Q9400 Quad-Core (6MB L2 cache,2.66GHz,1333FSB), ATI Radeon HD 2400 PRO 128MB, 4GB Dual Channel DDR2 SDRAM....

Interesting that you haven't seen this with exhaustive optimizer -- I have not looked at that. I have strategies with numerous input variables so need the genetic optimizer to be able to search any sort of reasonable input ranges.

geektrader
Posts: 100
Joined: 17 Jul 2009
Location: Germany

Postby geektrader » 18 Jul 2009

Yes, I also only use the Genetic Optimizer since my Neural Network Strategy has heaps of inputs. Anyway, you can try this on a smaller-input strategy and compare it. The "Exhausive Search" always uses all 4 Core´s from the start to the end of Optimization, only the "Genetic Optimizer" is having this problem. However, I also only need the Genetic Optimizer since I can´t run 1 trillion combinations in "Exhausive Search":) So hope they have a fix for that.

Cheers.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 20 Jul 2009

I confirm that we have an issue with Incremental genetic optimization and should improve the algorithm that doesn't use all cores currently.

However if you use Basic method the optimization will use all available cores.

geektrader
Posts: 100
Joined: 17 Jul 2009
Location: Germany

Postby geektrader » 20 Jul 2009

Great to hear that this will be worked on. Basic Optimization is not an option for me as I have combinations in the Trillions range;) Thanks Andrew.

SiliconTiger
Posts: 24
Joined: 21 Feb 2009

Postby SiliconTiger » 20 Jul 2009

Thanks Andrew for the confirmation this is a problem. I would also strongly urge you all to prioritize fixing this issue as most modern strategies have a reasonable number of inputs with input ranges which, in combination, far exceed what is possible to run using "exhaustive search". Genetic search is our only option.

Thanks again for your help and great support.

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 20 Jul 2009

I'm not sure why Basic method is not an option for you? It is just one of GA optimizer methods and it can work with many combinations too.

SiliconTiger
Posts: 24
Joined: 21 Feb 2009

Postby SiliconTiger » 20 Jul 2009

Andrew-- I have not altered the default settings for the generic algorithm (screenshot of the default settings I am using is attached), so the problem I am having is with the "basic" optimization method. I have not tried the incremental optimization method.

Please confirm you can reproduce this problem with the "basic" genetic optimization as well.

Thanks.
Attachments
Genetic Algorithm Properties.jpg
Genetic Algorithm Properties.jpg (45.33 KiB) Viewed 244 times

User avatar
Andrew Kirillov
Posts: 1589
Joined: 28 Jul 2005
Has thanked: 2 times
Been thanked: 31 times
Contact:

Postby Andrew Kirillov » 21 Jul 2009

It uses all 4 cores on our end. Could you contact our helpdesk for further investigation?


Return to “MultiCharts”