Strategy Tester not running at full speed.

I have a quad-core Q6600 machine which I use for Strategy Tester. I used to run four instances of Strategy Testers and all four will run at full speed (i.e. the CPU load will be at 100%).

Recently, I can no longer run four STs anymore without the CPU breaking off full speed. I can only run at most three instances of STs now. When I run three, the CPU load will be at 75%, but as soon as I power up the fourth, the CPU load will drop down to 10% only and the STs will take ages to finish.

Do you think the cut-off is because CPU safety feature? Is there any settings in BIOS that I tweak to prevent the CPU from breaking off?

What's the RAM usage?

My machine have 4GB RAM which I think is a plenty. Will the ST uses a lot of RAM if the history data goes back many years? Usually I only run the ST on a sample size of two to four years only.

Depends on the density of the data you're using. Rerun your tests and see how much ram is being used. What I think is happening is you are hitting your RAM limit, and the computer is switching to file swapping (using your HD as 'extended ram')

Thanks for the tip, Ronald. I will check this out.

Do you know how to make the ST uses multiple cores? If I run only 1 ST, it will only use one core which is kinda waste. Intel is producing 6 cores and 8 cores soon - it'll be a shame to have these on idle.

Proxmox.

Should allow you to combine cores virtually.

How should this proxmox magically parallelize single threaded code? Does it contain a time machine? If A needs to be finished before B can be started in a single threaded program then no automated voodoo, no matter how sophisticated, can break the laws of causality and the nature of space and time as we know them and make A and B start at the same time and still let B depend on A's outcome.

As I understand proxmox, it takes a set of resources (CPUs, Ram, HD space) and creates virtual instances similar to VMware.

But this can not parallelize the code. It can distribute a LOT of processes and threads optimally across the available CPUs, but one process that has only one thread can only run on one CPU. You simply cannot make 8 Cores run the same single threadd program at 8 times the speed.

The program itself must be written in a way that it can split up the workload into equally sized chunks that can be processed completely independent from each other at the same time before it starts. This is not a simple task! It needs a completely different approach to the problem solution, many programmers have never learned this and some problems simply cannot be splitted.

Modern CPUs can to some degree execute pieces of code that are independent in parallel without special measures and if there are only two possible outcomes at some point they can do such "dirty" tricks like branch prediction and just guess the outcome and start executing some later code and throw it away if the guess was wrong (and also you can have your compiler produce code that tends to execute the else branch much more often to support the predictive power), but this can not work if the possible outcomes of an earlier task are too many.

A real-world example: If you have 8 cars instead of one you cant transport a person from A to B any faster than with only one. You can transport 100 people from A to B eight times faster with eight cars but not one single person. No matter how sophisticated the algorithm is that choses which car to use.