amirouche
2018-03-21 00:14:56 UTC
What I want is that my CPUs to be always busy with the most important
work.
I don't know how CPU intensive or whether they block on system calls in
advance.
There is different kind of them. Some do IO, some do CPU.
I have a system where I submit a problem to a system with multiple
agents.
Each agent will concurrently have a try at the problem creating
theories.
The agent are coded using some greedy lazy algorithm. That is they don't
compute the whole solution and try to stream the result invocation after
invocation. So that, if a new theory is added by an agent, it might
delock
another agent or make it's computation faster. That is agents don't
block
for a long time.
Similarly when the user input some fact in the database, it might
simplify
the problem and the agents must re-evalute their strategy, maybe
thrashing
old theories.
The problem is complete, when all agents have no more theory to add to
the
problem.
Another problem, is that there is some backgound tasks that must always
be
executed when the whole system is idle.
Another feature, is that if the program crash for some reason, the
agents
must restart from where they left. agents are idempotent that is there
is no problem in running them twice in a row or after a crash. But, does
this mean I need to keep some kind of running tasks journal, to
be able to restart them?
I read, they are some event loops, that implements some kind of priority
system. In particular, the priority of each thread/fiber might change
over
time. How can this be achieved in guile? fibers?
I understand my explanation is not very good. I will try to improve
tomorrow.
Thanks in advance!
Amirouche ~ amz3 ~ http://hyperdev.fr
work.
I don't know how CPU intensive or whether they block on system calls in
advance.
There is different kind of them. Some do IO, some do CPU.
I have a system where I submit a problem to a system with multiple
agents.
Each agent will concurrently have a try at the problem creating
theories.
The agent are coded using some greedy lazy algorithm. That is they don't
compute the whole solution and try to stream the result invocation after
invocation. So that, if a new theory is added by an agent, it might
delock
another agent or make it's computation faster. That is agents don't
block
for a long time.
Similarly when the user input some fact in the database, it might
simplify
the problem and the agents must re-evalute their strategy, maybe
thrashing
old theories.
The problem is complete, when all agents have no more theory to add to
the
problem.
Another problem, is that there is some backgound tasks that must always
be
executed when the whole system is idle.
Another feature, is that if the program crash for some reason, the
agents
must restart from where they left. agents are idempotent that is there
is no problem in running them twice in a row or after a crash. But, does
this mean I need to keep some kind of running tasks journal, to
be able to restart them?
I read, they are some event loops, that implements some kind of priority
system. In particular, the priority of each thread/fiber might change
over
time. How can this be achieved in guile? fibers?
I understand my explanation is not very good. I will try to improve
tomorrow.
Thanks in advance!
Amirouche ~ amz3 ~ http://hyperdev.fr