doc/html/overview.shtml - SchedMD/slurm - Git at Google

 <!--#include virtual="header.txt"-->

 <h1><a name="top">Overview</a></h1>

 <p>Slurm is an open source,
 fault-tolerant, and highly scalable cluster management and job scheduling system
 for large and small Linux clusters. Slurm requires no kernel modifications for
 its operation and is relatively self-contained. As a cluster workload manager,
 Slurm has three key functions. First, it allocates exclusive and/or non-exclusive
 access to resources (compute nodes) to users for some duration of time so they
 can perform work. Second, it provides a framework for starting, executing, and
 monitoring work (normally a parallel job) on the set of allocated nodes.
 Finally, it arbitrates contention for resources by managing a queue of
 pending work.
 Optional plugins can be used for
 <a href="accounting.html">accounting</a>,
 <a href="reservations.html">advanced reservation</a>,
 <a href="gang_scheduling.html">gang scheduling</a> (time sharing for
 parallel jobs), backfill scheduling,
 <a href="topology.html">topology optimized resource selection</a>,
 <a href="resource_limits.html">resource limits</a> by user or account,
 and sophisticated <a href="priority_multifactor.html"> multifactor job
 prioritization</a> algorithms.

 <h2 id="architecture">Architecture
 <a class="slurm_link" href="#architecture"></a>
 </h2>
 <p>Slurm has a centralized manager, <b>slurmctld</b>, to monitor resources and
 work. There may also be a backup manager to assume those responsibilities in the
 event of failure. Each compute server (node) has a <b>slurmd</b> daemon, which
 can be compared to a remote shell: it waits for work, executes that work, returns
 status, and waits for more work.
 The <b>slurmd</b> daemons provide fault-tolerant hierarchical communications.
 There is an optional <b>slurmdbd</b> (Slurm DataBase Daemon) which can be used
 to record accounting information for multiple Slurm-managed clusters in a
 single database.
 There is an optional
 <a href="rest.html"><b>slurmrestd</b> (Slurm REST API Daemon)</a>
 which can be used to interact with Slurm through its
 <a href="https://en.wikipedia.org/wiki/Representational_state_transfer">
 REST API</a>.
 User tools include <b>srun</b> to initiate jobs,
 <b>scancel</b> to terminate queued or running jobs,
 <b>sinfo</b> to report system status,
 <b>squeue</b> to report the status of jobs, and
 <b>sacct</b> to get information about jobs and job steps that are running or have completed.
 The <b>sview</b> commands graphically reports system and
 job status including network topology.
 There is an administrative tool <b>scontrol</b> available to monitor
 and/or modify configuration and state information on the cluster.
 The administrative tool used to manage the database is <b>sacctmgr</b>.
 It can be used to identify the clusters, valid users, valid accounts, etc.
 APIs are available for all functions.</p>

 <div class="figure">
   <img src="arch.gif" width="550"><br>
   Figure 1. Slurm components
 </div>

 <p>Slurm has a general-purpose plugin mechanism available to easily support various
 infrastructures. This permits a wide variety of Slurm configurations using a
 building block approach. These plugins presently include:
 <ul>
 <li>Accounting Storage:
   Primarily Used to store historical data about jobs.  When used with
   SlurmDBD (Slurm Database Daemon), it can also supply a
   limits based system along with historical system status.
 </li>

 <li>Account Gather Energy:
   Gather energy consumption data per job or nodes in the system.
   This plugin is integrated with the
   Accounting Storage and Job Account Gather plugins.
 </li>

 <li>Authentication of communications:
   Provides authentication mechanism between various components of Slurm.
 </li>

 <li><a href="containers.html">Containers</a>:
   HPC workload container support and implementations.
 </li>

 <li>Credential (Digital Signature Generation):
   Mechanism used to generate a digital signature, which is used to validate
   that job step is authorized to execute on specific nodes.
   This is distinct from the plugin used for
   Authentication since the job step
   request is sent from the user's srun command rather than directly from the
   slurmctld daemon, which generates the job step credential and its
   digital signature.
 </li>

 <li><a href="gres.html">Generic Resources</a>: Provide interface to
   control generic resources, including Graphical Processing Units (GPUs).
 </li>

 <li><a href="job_submit_plugins.html">Job Submit</a>:
   Custom plugin to allow site specific control over job requirements at
   submission and update.
 </li>

 <li>Job Accounting Gather:
   Gather job step resource utilization data.
 </li>

 <li>Job Completion Logging:
   Log a job's termination data. This is typically a subset of data stored by
   an Accounting Storage Plugin.
 </li>

 <li>Launchers:
   Controls the mechanism used by the <a href="srun.html">'srun'</a> command
   to launch the tasks.
 </li>

 <li>MPI:
   Provides different hooks for the various MPI implementations.
   For example, this can set MPI specific environment variables.
 </li>

 <li><a href="preempt.html">Preempt</a>:
   Determines which jobs can preempt other jobs and the preemption mechanism
   to be used.
 </li>

 <li>Priority:
   Assigns priorities to jobs upon submission and on an ongoing basis
   (e.g. as they age).
 </li>

 <li>Process tracking (for signaling):
   Provides a mechanism for identifying the processes associated with each job.
   Used for job accounting and signaling.
 </li>

 <li>Scheduler:
   Plugin determines how and when Slurm schedules jobs.
 </li>

 <li>Node selection:
   Plugin used to determine the resources used for a job allocation.
 </li>

 <li><a href="site_factor.html">Site Factor (Priority)</a>:
   Assigns a specific site_factor component of a job's multifactor priority to
   jobs upon submission and on an ongoing basis (e.g. as they age).
 </li>

 <li>Switch or interconnect:
   Plugin to interface with a switch or interconnect.
   For most systems (Ethernet or InfiniBand) this is not needed.
 </li>

 <li>Task Affinity:
   Provides mechanism to bind a job and its individual tasks to specific
   processors.
 </li>

 <li>Network Topology:
   Optimizes resource selection based upon the network topology.
   Used for both job allocations and advanced reservation.
 </li>

 </ul>

 <p>The entities managed by these Slurm daemons, shown in Figure 2, include <b>nodes</b>,
 the compute resource in Slurm, <b>partitions</b>, which group nodes into logical
 sets, <b>jobs</b>, or allocations of resources assigned to a user for
 a specified amount of time, and <b>job steps</b>, which are sets of (possibly
 parallel) tasks within a job.
 The partitions can be considered job queues, each of which has an assortment of
 constraints such as job size limit, job time limit, users permitted to use it, etc.
 Priority-ordered jobs are allocated nodes within a partition until the resources
 (nodes, processors, memory, etc.) within that partition are exhausted. Once
 a job is assigned a set of nodes, the user is able to initiate parallel work in
 the form of job steps in any configuration within the allocation. For instance,
 a single job step may be started that utilizes all nodes allocated to the job,
 or several job steps may independently use a portion of the allocation.
 Slurm provides resource management for the processors allocated to a job,
 so that multiple job steps can be simultaneously submitted and queued until
 there are available resources within the job's allocation.</p>

 <div class="figure">
   <img src="entities.gif" width="550"><br>
   Figure 2. Slurm entities
 </div>


 <h2 id="configurability">Configurability
 <a class="slurm_link" href="#configurability"></a>
 </h2>
 <p>Node state monitored include: count of processors, size of real memory, size
 of temporary disk space, and state (UP, DOWN, etc.). Additional node information
 includes weight (preference in being allocated work) and features (arbitrary information
 such as processor speed or type).
 Nodes are grouped into partitions, which may contain overlapping nodes so they are
 best thought of as job queues.
 Partition information includes: name, list of associated nodes, state (UP or DOWN),
 maximum job time limit, maximum node count per job, group access list,
 priority (important if nodes are in multiple partitions) and shared node access policy
 with optional over-subscription level for gang scheduling (e.g. YES, NO or FORCE:2).
 Bit maps are used to represent nodes and scheduling
 decisions can be made by performing a small number of comparisons and a series
 of fast bit map manipulations. A sample (partial. Slurm configuration file follows.</p>
 <pre>
 #
 # Sample /etc/slurm.conf
 #
 SlurmctldHost=linux0001  # Primary server
 SlurmctldHost=linux0002  # Backup server
 #
 AuthType=auth/munge
 Epilog=/usr/local/slurm/sbin/epilog
 PluginDir=/usr/local/slurm/lib
 Prolog=/usr/local/slurm/sbin/prolog
 SlurmctldPort=7002
 SlurmctldTimeout=120
 SlurmdPort=7003
 SlurmdSpoolDir=/var/tmp/slurmd.spool
 SlurmdTimeout=120
 StateSaveLocation=/usr/local/slurm/slurm.state
 TmpFS=/tmp
 #
 # Node Configurations
 #
 NodeName=DEFAULT CPUs=4 TmpDisk=16384 State=IDLE
 NodeName=lx[0001-0002] State=DRAINED
 NodeName=lx[0003-8000] RealMemory=2048 Weight=2
 NodeName=lx[8001-9999] RealMemory=4096 Weight=6 Feature=video
 #
 # Partition Configurations
 #
 PartitionName=DEFAULT MaxTime=30 MaxNodes=2
 PartitionName=login Nodes=lx[0001-0002] State=DOWN
 PartitionName=debug Nodes=lx[0003-0030] State=UP Default=YES
 PartitionName=class Nodes=lx[0031-0040] AllowGroups=students
 PartitionName=DEFAULT MaxTime=UNLIMITED MaxNodes=4096
 PartitionName=batch Nodes=lx[0041-9999]
 </pre>

 <p style="text-align:center;">Last modified 6 August 2021</p>

 <!--#include virtual="footer.txt"-->
	<!--#include virtual="header.txt"-->

	<h1><a name="top">Overview</a></h1>

	<p>Slurm is an open source,
	fault-tolerant, and highly scalable cluster management and job scheduling system
	for large and small Linux clusters. Slurm requires no kernel modifications for
	its operation and is relatively self-contained. As a cluster workload manager,
	Slurm has three key functions. First, it allocates exclusive and/or non-exclusive
	access to resources (compute nodes) to users for some duration of time so they
	can perform work. Second, it provides a framework for starting, executing, and
	monitoring work (normally a parallel job) on the set of allocated nodes.
	Finally, it arbitrates contention for resources by managing a queue of
	pending work.
	Optional plugins can be used for
	<a href="accounting.html">accounting</a>,
	<a href="reservations.html">advanced reservation</a>,
	<a href="gang_scheduling.html">gang scheduling</a> (time sharing for
	parallel jobs), backfill scheduling,
	<a href="topology.html">topology optimized resource selection</a>,
	<a href="resource_limits.html">resource limits</a> by user or account,
	and sophisticated <a href="priority_multifactor.html"> multifactor job
	prioritization</a> algorithms.

	<h2 id="architecture">Architecture
	<a class="slurm_link" href="#architecture"></a>
	</h2>
	<p>Slurm has a centralized manager, <b>slurmctld</b>, to monitor resources and
	work. There may also be a backup manager to assume those responsibilities in the
	event of failure. Each compute server (node) has a <b>slurmd</b> daemon, which
	can be compared to a remote shell: it waits for work, executes that work, returns
	status, and waits for more work.
	The <b>slurmd</b> daemons provide fault-tolerant hierarchical communications.
	There is an optional <b>slurmdbd</b> (Slurm DataBase Daemon) which can be used
	to record accounting information for multiple Slurm-managed clusters in a
	single database.
	There is an optional
	<a href="rest.html"><b>slurmrestd</b> (Slurm REST API Daemon)</a>
	which can be used to interact with Slurm through its
	<a href="https://en.wikipedia.org/wiki/Representational_state_transfer">
	REST API</a>.
	User tools include <b>srun</b> to initiate jobs,
	<b>scancel</b> to terminate queued or running jobs,
	<b>sinfo</b> to report system status,
	<b>squeue</b> to report the status of jobs, and
	<b>sacct</b> to get information about jobs and job steps that are running or have completed.
	The <b>sview</b> commands graphically reports system and
	job status including network topology.
	There is an administrative tool <b>scontrol</b> available to monitor
	and/or modify configuration and state information on the cluster.
	The administrative tool used to manage the database is <b>sacctmgr</b>.
	It can be used to identify the clusters, valid users, valid accounts, etc.
	APIs are available for all functions.</p>

	<div class="figure">
	<img src="arch.gif" width="550"><br>
	Figure 1. Slurm components
	</div>

	<p>Slurm has a general-purpose plugin mechanism available to easily support various
	infrastructures. This permits a wide variety of Slurm configurations using a
	building block approach. These plugins presently include:
	<ul>
	<li>Accounting Storage:
	Primarily Used to store historical data about jobs. When used with
	SlurmDBD (Slurm Database Daemon), it can also supply a
	limits based system along with historical system status.
	</li>

	<li>Account Gather Energy:
	Gather energy consumption data per job or nodes in the system.
	This plugin is integrated with the
	Accounting Storage and Job Account Gather plugins.
	</li>

	<li>Authentication of communications:
	Provides authentication mechanism between various components of Slurm.
	</li>

	<li><a href="containers.html">Containers</a>:
	HPC workload container support and implementations.
	</li>

	<li>Credential (Digital Signature Generation):
	Mechanism used to generate a digital signature, which is used to validate
	that job step is authorized to execute on specific nodes.
	This is distinct from the plugin used for
	Authentication since the job step
	request is sent from the user's srun command rather than directly from the
	slurmctld daemon, which generates the job step credential and its
	digital signature.
	</li>

	<li><a href="gres.html">Generic Resources</a>: Provide interface to
	control generic resources, including Graphical Processing Units (GPUs).
	</li>

	<li><a href="job_submit_plugins.html">Job Submit</a>:
	Custom plugin to allow site specific control over job requirements at
	submission and update.
	</li>

	<li>Job Accounting Gather:
	Gather job step resource utilization data.
	</li>

	<li>Job Completion Logging:
	Log a job's termination data. This is typically a subset of data stored by
	an Accounting Storage Plugin.
	</li>

	<li>Launchers:
	Controls the mechanism used by the <a href="srun.html">'srun'</a> command
	to launch the tasks.
	</li>

	<li>MPI:
	Provides different hooks for the various MPI implementations.
	For example, this can set MPI specific environment variables.
	</li>

	<li><a href="preempt.html">Preempt</a>:
	Determines which jobs can preempt other jobs and the preemption mechanism
	to be used.
	</li>

	<li>Priority:
	Assigns priorities to jobs upon submission and on an ongoing basis
	(e.g. as they age).
	</li>

	<li>Process tracking (for signaling):
	Provides a mechanism for identifying the processes associated with each job.
	Used for job accounting and signaling.
	</li>

	<li>Scheduler:
	Plugin determines how and when Slurm schedules jobs.
	</li>

	<li>Node selection:
	Plugin used to determine the resources used for a job allocation.
	</li>

	<li><a href="site_factor.html">Site Factor (Priority)</a>:
	Assigns a specific site_factor component of a job's multifactor priority to
	jobs upon submission and on an ongoing basis (e.g. as they age).
	</li>

	<li>Switch or interconnect:
	Plugin to interface with a switch or interconnect.
	For most systems (Ethernet or InfiniBand) this is not needed.
	</li>

	<li>Task Affinity:
	Provides mechanism to bind a job and its individual tasks to specific
	processors.
	</li>

	<li>Network Topology:
	Optimizes resource selection based upon the network topology.
	Used for both job allocations and advanced reservation.
	</li>

	</ul>

	<p>The entities managed by these Slurm daemons, shown in Figure 2, include <b>nodes</b>,
	the compute resource in Slurm, <b>partitions</b>, which group nodes into logical
	sets, <b>jobs</b>, or allocations of resources assigned to a user for
	a specified amount of time, and <b>job steps</b>, which are sets of (possibly
	parallel) tasks within a job.
	The partitions can be considered job queues, each of which has an assortment of
	constraints such as job size limit, job time limit, users permitted to use it, etc.
	Priority-ordered jobs are allocated nodes within a partition until the resources
	(nodes, processors, memory, etc.) within that partition are exhausted. Once
	a job is assigned a set of nodes, the user is able to initiate parallel work in
	the form of job steps in any configuration within the allocation. For instance,
	a single job step may be started that utilizes all nodes allocated to the job,
	or several job steps may independently use a portion of the allocation.
	Slurm provides resource management for the processors allocated to a job,
	so that multiple job steps can be simultaneously submitted and queued until
	there are available resources within the job's allocation.</p>

	<div class="figure">
	<img src="entities.gif" width="550"><br>
	Figure 2. Slurm entities
	</div>


	<h2 id="configurability">Configurability
	<a class="slurm_link" href="#configurability"></a>
	</h2>
	<p>Node state monitored include: count of processors, size of real memory, size
	of temporary disk space, and state (UP, DOWN, etc.). Additional node information
	includes weight (preference in being allocated work) and features (arbitrary information
	such as processor speed or type).
	Nodes are grouped into partitions, which may contain overlapping nodes so they are
	best thought of as job queues.
	Partition information includes: name, list of associated nodes, state (UP or DOWN),
	maximum job time limit, maximum node count per job, group access list,
	priority (important if nodes are in multiple partitions) and shared node access policy
	with optional over-subscription level for gang scheduling (e.g. YES, NO or FORCE:2).
	Bit maps are used to represent nodes and scheduling
	decisions can be made by performing a small number of comparisons and a series
	of fast bit map manipulations. A sample (partial. Slurm configuration file follows.</p>
	<pre>
	#
	# Sample /etc/slurm.conf
	#
	SlurmctldHost=linux0001 # Primary server
	SlurmctldHost=linux0002 # Backup server
	#
	AuthType=auth/munge
	Epilog=/usr/local/slurm/sbin/epilog
	PluginDir=/usr/local/slurm/lib
	Prolog=/usr/local/slurm/sbin/prolog
	SlurmctldPort=7002
	SlurmctldTimeout=120
	SlurmdPort=7003
	SlurmdSpoolDir=/var/tmp/slurmd.spool
	SlurmdTimeout=120
	StateSaveLocation=/usr/local/slurm/slurm.state
	TmpFS=/tmp
	#
	# Node Configurations
	#
	NodeName=DEFAULT CPUs=4 TmpDisk=16384 State=IDLE
	NodeName=lx[0001-0002] State=DRAINED
	NodeName=lx[0003-8000] RealMemory=2048 Weight=2
	NodeName=lx[8001-9999] RealMemory=4096 Weight=6 Feature=video
	#
	# Partition Configurations
	#
	PartitionName=DEFAULT MaxTime=30 MaxNodes=2
	PartitionName=login Nodes=lx[0001-0002] State=DOWN
	PartitionName=debug Nodes=lx[0003-0030] State=UP Default=YES
	PartitionName=class Nodes=lx[0031-0040] AllowGroups=students
	PartitionName=DEFAULT MaxTime=UNLIMITED MaxNodes=4096
	PartitionName=batch Nodes=lx[0041-9999]
	</pre>

	<p style="text-align:center;">Last modified 6 August 2021</p>

	<!--#include virtual="footer.txt"-->