dragon.native.machine

The Dragon native machine interface provides hardware related functionality.

These calls return various hardware properties of the system(s), like the number of processors or a list of all nodes.

Functions

`cpu_count`()	Return the number of CPUs in this system.
`current`()	Return a node object for this process.

Classes

`Node`
`System`	System abstraction representing the collection of nodes in the Dragon runtime.

cpu_count() → int[source]

Return the number of CPUs in this system.

Returns:: integer representing the number of cpus

class Node[source]

Bases: object

__init__(ident: str)[source]

A class abstracting the current hardware node.

Parameters:: ident (str or int) – name or h_uid of the node to request.

property name: str

Return the Dragon name of this node. Use Node.hostname for the hostname.

Returns:: The name of the node
Return type:: str

property h_uid: int

The unique global identifier of this node.

Returns:: the h_uid
Return type:: int

property ip: str

Return the IPv4 address of this node

Returns:: The addres
Return type:: str

property num_cpus: int

Return the number of CPUs of this node.

Returns:: The number of CPUs
Return type:: int

property num_gpus: int

Return a the number of GPUs on this node

Returns:: The number of GPUs
Return type:: list[int]

property physical_mem: int

Return the physical memory of this node.

Returns:: Physical memory in bytes
Return type:: int

property gpus: list[int]

Return a list of GPU visible devices on this node

Returns:: list of GPU visible devices
Return type:: list[int]

property gpu_vendor: str

Return the name of the GPU Vendor on this node

Returns:: GPU vendor name
Return type:: str

property gpu_env_str: str

Return the environment variable used by GPU Vendor to define GPU affinity

Returns:: GPU environment variable string
Return type:: str

property cpus: list[int]

Return the CPUs available on this node

Returns:: list of CPUs
Return type:: list[int]

property hostname: str

Host name of the node in the network.

This is not the Dragon internal name (Node.name) for this node.

Returns:: The hostname
Return type:: str

property is_primary: bool

Returns true if this is the primary node for the runtime. The primary node is important in that it is the node with global services.

Returns:: is the primary node
Return type:: bool

current() → Node[source]

Return a node object for this process.

current() can be imported from the top-level Dragon or dragon.native.machine modules.

from dragon import current

or

from dragon.native.machine import current

Returns:: Node object
Return type:: Node

class System[source]

Bases: object

System abstraction representing the collection of nodes in the Dragon runtime. Provides access to node information, primary node, and policies for CPU and GPU placement.

System can be imported from the top-level Dragon or dragon.native.machine modules.

from dragon import System

or

from dragon.native.machine import System

__init__()[source]: A stub of a system abstraction

property nodes

property nnodes: int

property primary_node: Node

property restarted: bool

Helper function to check environment and determine if the runtime was previously restarted

Returns:: true if the runtime was restarted
Return type:: bool

hostname_policies() → list[Policy][source]

Generate a list of policies that use all of the nodes in the allocation specified by hostname

Returns:: a list of policies where each policy specifies hostname placement and the hostname, without duplication
Return type:: list[Policy]

gpu_policies() → list[Policy][source]

Generate a list of policies that use all of the GPUs in the allocation

Returns:: a list of policies where each policy specifies hostname placement, the hostname, and a gpu on the node, without duplication
Return type:: list

static cpu_count() → int[source]

Return the number of CPUs in this system.

Returns:: integer representing the number of cpus
Return type:: int