assuncaomarcos / resource-availability Goto Github PK

This Python library provides a data structure, termed availability profile, for managing the availability of computing resources. The structure is handy for simulations and experiments where one must track the compute cluster resources allocated to jobs or tasks over time. The following provides examples of using the discrete resource range, set, and profile, which use int as data type, but the same concepts apply to profiles for other data types.

One can use the discrete (int) range, set, and profile to track the availability of, for instance, CPUs or cluster nodes. To create ranges with resources 0..20 and 30..50 and add them to a set:

from availability.sets import DiscreteRange, DiscreteSet
span1 = DiscreteRange(0, 20)
span2 = DiscreteRange(30, 50)
res_set = DiscreteSet([span1, span2])

Although you can create ranges and sets, one will not manipulate them directly. For tracking the resources available over time, one will likely use an availability profile (discrete or continuous, depending on the type of resource they are dealing with). To create an availability profile with a maximum capacity of 100 discrete resources for tracking the availability of cluster nodes, for instance, one can use the following:

from availability.profile import DiscreteProfile
profile = DiscreteProfile(max_capacity=100)

If you are using the profile in a task-scheduling simulation, you can use the method allocate_resources() from the profile to remove the resource range 0..10 assigned to the task:

profile.allocate_resources(
    resources=DiscreteSet(
        [DiscreteRange(0, 10)]
    ),
    start_time=0,
    end_time=10
)

To find the time at which a task requiring 40 resources for 50 time units can start:

slot = profile.find_start_time(
    quantity=40, ready_time=5, duration=50
)

The returned slot will resemble:

TimeSlot(
    start_time=0,
    end_time=50,
    resources=DiscreteSet([DiscreteRange(10, 100)])
)

The profile provides other methods, such as check_availability() to check whether a given quantity of resources is available over a given period:

slot = profile.check_availability(
    quantity=10, start_time=5, duration=50
)

One can use the methods free_time_slots() or scheduling_options() to obtain the list of time slots and resources available. The main difference between them is that the time slots returned by the latter may overlap as they represent all the scheduling possibilities for scheduling a job, given the resource availability over the specified period:

slots = profile.scheduling_options(
    start_time=10,
    end_time=100,
    min_duration=20,
    min_quantity=5
)

The operations for querying the resources available during a period return the complete set of resources available. This design allows a user to implement their resource selection policy. However, you can use select_resources() or select_slot_resources() to select a given number of resources from a set or slot:

slot = profile.find_start_time(
    quantity=5, ready_time=0, duration=10
)
selected = profile.select_resources(
    resources=slot.resources, quantity=5)
)