Public discussion of the formula for calculating the cost of resources in the MQL5 Cloud Network - page 22
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Count the number of agents.
Raw thoughts... something like this may have been flashed before...
A buyer of processor time sends an application to some MQ service. Let's call it the "bid placement stage".
At this stage, the "buyer's" expert is run once on the MQ equipment. This test will give some sort of "rating" of resource intensity of the Expert Advisor (roughly speaking, how much time is required for the Expert Advisor to completely pass one run on some "reference configuration/performance" computer). Here, of course, there is a nuance. Depending on the different parameters being optimised, the resource intensity can vary... and probably significantly.
The application is then published to some kind of technical exchange, which specifies:
1. resource intensity rating of the EA.
2. number of necessary runs.
3. Period of optimization + possibility of partial execution.
4. desired price (for one 1/10/100/1000/... runs).
Processor time sellers have two options of counter bid submission: automatic and manual.
With automatic bidding the seller sets the price to which he agrees. The system monitors the exchange itself and, if there is a matching bid, takes it into operation.
With manual bidding, the seller may take a job that pays less.
Another nuance.
For the buyer it is not only the cost but also the deadline. In fact, just like in life.
Therefore, the seller of his CPU time can submit a counter bid only for the amount of tasks that he can master within the time limit given by the client.
For example, the seller has a comp with 4 cores/agents with a performance rating of 1 unit each. Note: Each agent in the MQ service has a performance rating.
Next, the MQ service counts: can this seller complete the entire volume (number of runs) in the allotted time. If the buyer has indicated the possibility of partial performance, the latter is taken into processing at once. If partial performance is forbidden, MQ "looks" until there are enough sellers to complete the entire task (number of runs) within the specified time.
What remains unclear here is the issue of "splitting the pie" between sellers and monitoring the sellers' performance:
1. Clear cheaters.
2. irresponsible people who load their power with personal/other tasks, lowering the efficiency of each/any agent/processor core.
P.S. Apologies for the rambling. Just as a "think about it" information. :)
Developers: In agent manager you can select all agents by pressing CTRL+A (select all). Currently, to disable all of them, I have to select with shifft and then disable/enable them. I think when there are a lot of agents, this is very relevant.
Select, say, 5th core, then hold down Shit and click on the 7th core
And everything is selectable normally.
you can select the 1st core and move to the 8th core by holding down the shift ,
Do a click on the 8th core
Select let's say the 5th core then hold down Shit and click on the 7th core
and everything is fine
You can switch to the 8th core by holding down shift,
go straight to 8 cores
Thank you, I understand that very well. I mean, if I have 20-30 cores, it is faster/easier to select all of them with the left hand, as it is usually done in most applications, and use the right mouse to enable/disable them. Can be added to the context menu item "Select all". A trifle, but it would be more pleasant. Hotkeys always accelerate the work and improve the interface. I have no problem turning on agents in my spare time, but when I'm at my computer I turn everything off. I've had it on your network for two days, it's the only thing I noticed immediately. For example I work for half an hour by myself, half an hour counting agents, then me again, etc. Every time with the shifter is a bit annoying. How about a button "on all" and "off all"?
It is suggested to take a few parameters (for each agent individually) to start the discussion:
...
PR - agent performance index (unreliable value, falsifiable)
By the way, how do I get the PR value for my agents now? I only found information that PR should be automatically calculated when I start optimising an EA. I ran optimisation, tried 100 combinations using the slow method - PR did not appear.
From another build.
Thanks for the link... So is this the only way (it will be) now? Is there no way to find out the PR locally?