Reducing CPU load of NinjaTrader

August 15th, 2011, 08:08 AM

What I do for some of indicators is, I skip some cpu intensive logic if appropriate, when new bars start printing less then 1-3 seconds apart. In some indicators I only call draw methods or certain calculation methods, only when the chart is scrolled horizontally, or the scale changes vertically, or sometimes only if the chart is visible, depending on what the code does.

August 15th, 2011, 05:51 PM

gkinlaw

Here's an image of the task mgr with 2 cpu's running, I get spike's now instead of capping off at 25%. Also yesterday I found that I was recording data in Jan 1. and I cut that off. Bet that used up some cpu.

thanks for you replies!!

Your image did not come through. Click on the word screenshot in my post to learn how to attach images.

Mike

November 5th, 2011, 08:43 PM

bottom line DrawRegion is a CPU pig 256 bars or not

from Windows Resoure Monitor, CPU spike graphs ok, plateaued CPU mesa graphs bad

all I am doing is filling the area between +-2 tick offset of EMA. I ended up creating 2 new datasets to make use of DrawRegion.

I added a parm that if ture use DrawRegion if false use BackGround=color.touse

night and day difference....

while the filled regions looks kwel... I can not afford the CPU hit

to me boils down to form over function, I choose function.

November 6th, 2011, 05:03 AM

tulanch

bottom line DrawRegion is a CPU pig 256 bars or not

from Windows Resoure Monitor, CPU spike graphs ok, plateaued CPU mesa graphs bad

all I am doing is filling the area between +-2 tick offset of EMA. I ended up creating 2 new datasets to make use of DrawRegion.

I added a parm that if ture use DrawRegion if false use BackGround=color.touse

night and day difference....

while the filled regions looks kwel... I can not afford the CPU hit

to me boils down to form over function, I choose function.

Depends on how you coded it. The impact on the CPU depends on

-> how many regions you draw, if you draw from CurrentBar to CurrentBar - 1, this is the worst solution
-> whether you perform the calculations with each incoming tick or only with the first tick of a bar

November 6th, 2011, 04:53 PM

I'm doing bar close... not every tick...

ultimately goal is to fill area between 9ema 30wma red fill area when 30 on top, green if on bottom...pull back area reminder...

how can one do this other than current bar to current bar - 1 ?

perhaps keep growing the area till it changes ie current bar to current-n thus making 1 region instead of n regions?

other ideas/concepts?

November 6th, 2011, 06:54 PM

tulanch

I'm doing bar close... not every tick...

ultimately goal is to fill area between 9ema 30wma red fill area when 30 on top, green if on bottom...pull back area reminder...

how can one do this other than current bar to current bar - 1 ?

perhaps keep growing the area till it changes ie current bar to current-n thus making 1 region instead of n regions?

There is a coding example here, how to do it:

Download: Trigger Lines

Exported using NT 7.0.1000.15 This is a simple implementation of the trigger lines which can be used as a trend filter. The trigger lines are composed of a linear regression indicator (LinReg) and a signal line, which is calculated as the exponential …

Fat Tails

But then you should run it in CalculateOnBarClose = true mode. If you want to use the indicator in CalculateOnBarClose = false mode, you should probably select a displacement of 1.

November 7th, 2011, 09:44 AM

tulanch

bottom line DrawRegion is a CPU pig 256 bars or not

from Windows Resoure Monitor, CPU spike graphs ok, plateaued CPU mesa graphs bad

all I am doing is filling the area between +-2 tick offset of EMA. I ended up creating 2 new datasets to make use of DrawRegion.

I added a parm that if ture use DrawRegion if false use BackGround=color.touse

night and day difference....

while the filled regions looks kwel... I can not afford the CPU hit

to me boils down to form over function, I choose function.

Hi tulanch,

I use draw region a lot. Try changing historic load setting from days to bars and only load 200 to 500 bars.

Also, remember historic data stored in NT must be manually deleted.

Same is true with market replay.

If you have not done this cleanup in a while, if ever. Take the time to streamline NT.

That helps NT run without dragging all that old data around.

Lastly, Templates. If you have like 50 templates loaded, move or delete out the ones you are not using.

Templates may be running all the time even when when not loaded to charts.

This puts a huge load on the CPU.

Lastly, do a repair database in NT to compact data.

RJay

November 8th, 2011, 09:26 AM

WOW! huge, did I say HUGE difference.....

1.) had it measured in days, now using bars
2.) had 15 templates...removed all but the current

lightning fast now.... can not even tell I pressed F5

thanks again..

Reducing CPU load of NinjaTrader

Discussion in NinjaTrader

Reducing CPU load of NinjaTrader