Slurm merlin7 Configuration¶
This documentation shows basic Slurm configuration and options needed to run jobs in the Merlin7 cluster.
Public partitions configuration summary¶
CPU public partitions¶
| PartitionName | DefaultTime | MaxTime | Priority | Account | Per Job Limits | Per User Limits |
|---|---|---|---|---|---|---|
| general | 1-00:00:00 | 7-00:00:00 | Low | merlin | cpu=1024,mem=1920G | cpu=1024,mem=1920G |
| daily | 0-01:00:00 | 1-00:00:00 | Medium | merlin | cpu=1024,mem=1920G | cpu=2048,mem=3840G |
| hourly | 0-00:30:00 | 0-01:00:00 | High | merlin | cpu=2048,mem=3840G | cpu=8192,mem=15T |
| interactive | 0-04:00:00 | 0-12:00:00 | Highest | merlin | cpu=16,mem=30G,node=1 | cpu=32,mem=60G,node=1 |
GPU public partitions¶
A100 nodes¶
| PartitionName | DefaultTime | MaxTime | Priority | Account | Per Job Limits | Per User Limits |
|---|---|---|---|---|---|---|
| a100-general | 1-00:00:00 | 7-00:00:00 | Low | merlin | gres/gpu=4 | gres/gpu=8 |
| a100-daily | 0-01:00:00 | 1-00:00:00 | Medium | merlin | gres/gpu=8 | gres/gpu=8 |
| a100-hourly | 0-00:30:00 | 0-01:00:00 | High | merlin | gres/gpu=8 | gres/gpu=8 |
| a100-interactive | 0-01:00:00 | 0-12:00:00 | Very High | merlin | cpu=16,gres/gpu=1,mem=60G,node=1 | cpu=16,gres/gpu=1,mem=60G,node=1 |
Grace-Hopper nodes¶
| PartitionName | DefaultTime | MaxTime | Priority | Account | Per Job Limits | Per User Limits |
|---|---|---|---|---|---|---|
| gh-general | 1-00:00:00 | 7-00:00:00 | Low | merlin | gres/gpu=4 | gres/gpu=8 |
| gh-daily | 0-01:00:00 | 1-00:00:00 | Medium | merlin | gres/gpu=8 | gres/gpu=8 |
| gh-hourly | 0-00:30:00 | 0-01:00:00 | High | merlin | gres/gpu=8 | gres/gpu=8 |
| gh-interactive | 0-01:00:00 | 0-12:00:00 | Very High | merlin | cpu=16,gres/gpu=1,mem=46G,node=1 | cpu=16,gres/gpu=1,mem=46G,node=1 |
CPU cluster: merlin7¶
By default, jobs will be submitted to merlin7, as it is the primary cluster configured on the login nodes.
Specifying the cluster name is typically unnecessary unless you have defined environment variables that could override the default cluster name.
However, when necessary, one can specify the cluster as follows:
CPU general configuration¶
The Merlin7 CPU cluster is configured with the CR_CORE_MEMORY and CR_ONE_TASK_PER_CORE options.
- This configuration treats both cores and memory as consumable resources.
- Since the nodes are running with hyper-threading enabled, each core thread is counted as a CPU to fulfill a job's resource requirements.
By default, Slurm will allocate one task per core, which means: * Each task will consume 2 CPUs, regardless of whether both threads are actively used by the job.
This behavior ensures consistent resource allocation but may result in underutilization of hyper-threading in some cases.
CPU nodes definition¶
The table below provides an overview of the Slurm configuration for the different node types in the Merlin7 cluster. This information is essential for understanding how resources are allocated, enabling users to tailor their submission scripts accordingly.
| Nodes | Sockets | CoresPerSocket | Cores | ThreadsPerCore | CPUs | MaxMemPerNode | DefMemPerCPU | Features |
|---|---|---|---|---|---|---|---|---|
| login[001-002] | 2 | 64 | 128 | 2 | 256 | 480G | 1920M | AMD_EPYC_7713 |
| cn[001-077] | 2 | 64 | 128 | 2 | 256 | 480G | 1920M | AMD_EPYC_7713 |
Notes on memory configuration:
* Memory allocation options: To request additional memory, use the following options in your submission script:
* --mem=<mem_in_MB>: Allocates memory per node.
* --mem-per-cpu=<mem_in_MB>: Allocates memory per CPU (equivalent to a core thread).
The total memory requested cannot exceed the MaxMemPerNode value.
* Impact of disabling Hyper-Threading: Using the --hint=nomultithread option disables one thread per core,
effectively halving the number of available CPUs. Consequently, memory allocation will also be halved unless explicitly
adjusted.
For MPI-based jobs, where performance generally improves with single-threaded CPUs, this option is recommended.
In such cases, you should double the --mem-per-cpu value to account for the reduced number of threads.
Tip
Always verify the Slurm /var/spool/slurmd/conf-cache/slurm.conf configuration file for potential changes.
User and job limits with QoS¶
In the merlin7 CPU cluster, we enforce certain limits on jobs and users to ensure fair resource usage and prevent
overuse by a single user or job. These limits aim to balance resource availability while maintaining overall cluster
efficiency. However, applying limits can occasionally impact the cluster’s utilization. For example, user-specific
limits may result in pending jobs even when many nodes are idle due to low activity.
On the other hand, these limits also enhance cluster efficiency by preventing scenarios such as a single job monopolizing all available resources, which could block other jobs from running. Without job size limits, for instance, a large job might drain the entire cluster to satisfy its resource request, a situation that is generally undesirable.
Thus, setting appropriate limits is essential to maintain fair resource usage while optimizing cluster efficiency. These limits should allow for a mix of jobs of varying sizes and types, including single-core and parallel jobs, to coexist effectively.
To implement these limits, we utilize Quality of Service (QoS). Different QoS policies are defined and applied
to specific partitions in line with the established resource allocation policies. The table below outlines the
various QoS definitions applicable to the merlin7 CPU-based cluster. Here:
* MaxTRES specifies resource limits per job.
* MaxTRESPU specifies resource limits per user.
| Name | MaxTRES | MaxTRESPU | Scope |
|---|---|---|---|
| normal | partition | ||
| cpu_general | cpu=1024,mem=1920G | cpu=1024,mem=1920G | user, partition |
| cpu_daily | cpu=1024,mem=1920G | cpu=2048,mem=3840G | partition |
| cpu_hourly | cpu=2048,mem=3840G | cpu=8192,mem=15T | partition |
| cpu_interactive | cpu=16,mem=30G,node=1 | cpu=32,mem=60G,node=1 | partition |
Where:
* normal QoS: This QoS has no limits and is typically applied to partitions that do not require user or job
restrictions.
* cpu_general QoS: This is the default QoS for merlin7 users. It limits the total resources available to each
user. Additionally, this QoS is applied to the general partition, enforcing restrictions at the partition level and
overriding user-level QoS.
* cpu_daily QoS: Guarantees increased resources for the daily partition, accommodating shorter-duration jobs
with higher resource needs.
* cpu_hourly QoS: Offers the least constraints, allowing more resources to be used for the hourly partition,
which caters to very short-duration jobs.
* cpu_interactive QoS: Is restricted to one node and a few CPUs only, and is intended to be used when interactive
allocations are necessary (salloc, srun).
For additional details, refer to the CPU partitions section.
Tip
Always verify QoS definitions for potential changes using the sacctmgr show qos format="Name%22,MaxTRESPU%35,MaxTRES%35" command.
CPU partitions¶
This section provides a summary of the partitions available in the merlin7 CPU cluster.
Key concepts:
* PriorityJobFactor: This value is added to a job’s priority (visible in the PARTITION column of the sprio -l command).
Jobs submitted to partitions with higher PriorityJobFactor values generally run sooner. However, other factors like job age
and especially fair share can also influence scheduling.
* PriorityTier: Jobs submitted to partitions with higher PriorityTier values take precedence over pending jobs in partitions
with lower PriorityTier values. Additionally, jobs from higher PriorityTier partitions can preempt running jobs in lower-tier
partitions, where applicable.
* QoS: Specifies the quality of service associated with a partition. It is used to control and restrict resource availability
for specific partitions, ensuring that resource allocation aligns with intended usage policies. Detailed explanations of the various
QoS settings can be found in the User and job limits with QoS section.
Tip
Always verify partition configurations for potential changes using the scontrol show partition command.
CPU public partitions¶
| PartitionName | DefaultTime | MaxTime | TotalNodes | PriorityJobFactor | PriorityTier | QoS | AllowAccounts |
|---|---|---|---|---|---|---|---|
| general | 1-00:00:00 | 7-00:00:00 | 46 | 1 | 1 | cpu_general | merlin |
| daily | 0-01:00:00 | 1-00:00:00 | 58 | 500 | 1 | cpu_daily | merlin |
| hourly | 0-00:30:00 | 0-01:00:00 | 77 | 1000 | 1 | cpu_hourly | merlin |
| interactive | 0-04:00:00 | 0-12:00:00 | 58 | 1 | 2 | cpu_interactive | merlin |
All Merlin users are part of the merlin account, which is used as the default account when submitting jobs.
Similarly, if no partition is specified, jobs are automatically submitted to the general partition by default.
Tip
For jobs running less than one day, submit them to the daily partition. For jobs running less than one hour, use the hourly partition. These partitions provide higher priority and ensure quicker scheduling compared to general, which has limited node availability.
The hourly partition may include private nodes as an additional buffer. However, the current Slurm partition configuration, governed
by PriorityTier, ensures that jobs submitted to private partitions are prioritized and processed first. As a result, access to the
hourly partition might experience delays in such scenarios.
The interactive partition is designed specifically for real-time, interactive work. Here are the key characteristics:
- CPU Oversubscription: This partition allows CPU oversubscription (configured as
FORCE:4), meaning that up to four interactive jobs may share the same physical CPU core. This can impact performance, but enables fast access for short-term tasks. - Highest Scheduling Priority: Jobs submitted to the interactive partition are always prioritized. They will be scheduled before any jobs in other partitions.
- Intended Use: This partition is ideal for debugging, testing, compiling, short interactive runs, and other activities where immediate access is important.
Warning
Because of CPU sharing, the performance on the interactive partition may not be optimal for compute-intensive tasks. For long-running or production workloads, use a dedicated batch partition instead.
CPU private partitions¶
CAS / ASA¶
| PartitionName | DefaultTime | MaxTime | TotalNodes | PriorityJobFactor | PriorityTier | QoS | AllowAccounts |
|---|---|---|---|---|---|---|---|
| asa | 0-01:00:00 | 14-00:00:00 | 10 | 1 | 2 | normal | asa |
CNM / Mu3e¶
| PartitionName | DefaultTime | MaxTime | TotalNodes | PriorityJobFactor | PriorityTier | QoS | AllowAccounts |
|---|---|---|---|---|---|---|---|
| mu3e | 1-00:00:00 | 7-00:00:00 | 4 | 1 | 2 | normal | mu3e, meg |
CNM / MeG¶
| PartitionName | DefaultTime | MaxTime | TotalNodes | PriorityJobFactor | PriorityTier | QoS | AllowAccounts |
|---|---|---|---|---|---|---|---|
| meg-short | 0-01:00:00 | 0-01:00:00 | unlimited | 1000 | 2 | normal | meg |
| meg-long | 1-00:00:00 | 5-00:00:00 | unlimited | 1 | 2 | normal | meg |
| meg-prod | 1-00:00:00 | 5-00:00:00 | unlimited | 1000 | 4 | normal | meg |
GPU cluster: gmerlin7¶
As mentioned in previous sections, by default, jobs will be submitted to merlin7, as it is the primary cluster configured on the login nodes.
For submittng jobs to the GPU cluster, the cluster name gmerlin7 must be specified, as follows:
GPU general configuration¶
The Merlin7 GPU cluster is configured with the CR_CORE_MEMORY, CR_ONE_TASK_PER_CORE, and ENFORCE_BINDING_GRES options.
- This configuration treats both cores and memory as consumable resources.
- Since the nodes are running with hyper-threading enabled, each core thread is counted as a CPU to fulfill a job's resource requirements.
- Slurm will allocate the CPUs to the selected GPU.
By default, Slurm will allocate one task per core, which means:
- For hyper-threaded nodes (NVIDIA A100-based nodes), each task will consume 2 CPUs, regardless of whether both threads are actively used by the job.
- For the NVIDIA GraceHopper-based nodes, each task will consume 1 CPU.
This behavior ensures consistent resource allocation but may result in underutilization of hyper-threading in some cases.
GPU nodes definition¶
The table below provides an overview of the Slurm configuration for the different node types in the Merlin7 cluster. This information is essential for understanding how resources are allocated, enabling users to tailor their submission scripts accordingly.
| Nodes | Sockets | CoresPerSocket | Cores | ThreadsPerCore | CPUs | MaxMemPerNode | DefMemPerCPU | Gres | Features |
|---|---|---|---|---|---|---|---|---|---|
| gpu[001-007] | 4 | 72 | 288 | 1 | 288 | 828G | 2944M | gpu:gh200:4 | AMD_EPYC_7713, NV_A100 |
| gpu[101-105] | 1 | 64 | 64 | 2 | 128 | 480G | 3840M | gpu:nvidia_a100-sxm4-80gb:4 | GH200, NV_H100 |
Notes on memory configuration:
* Memory allocation options: To request additional memory, use the following options in your submission script:
* --mem=<mem_in_MB>: Allocates memory per node.
* --mem-per-cpu=<mem_in_MB>: Allocates memory per CPU (equivalent to a core thread).
The total memory requested cannot exceed the MaxMemPerNode value.
- Impact of disabling Hyper-Threading: Using the
--hint=nomultithreadoption disables one thread per core, effectively halving the number of available CPUs. Consequently, memory allocation will also be halved unless explicitly adjusted.
For MPI-based jobs, where performance generally improves with single-threaded CPUs, this option is recommended.
In such cases, you should double the --mem-per-cpu value to account for the reduced number of threads.
Tip
Always verify the Slurm /var/spool/slurmd/conf-cache/slurm.conf configuration file for potential changes.
User and job limits with QoS¶
In the gmerlin7 CPU cluster, we enforce certain limits on jobs and users to ensure fair resource usage and prevent
overuse by a single user or job. These limits aim to balance resource availability while maintaining overall cluster
efficiency. However, applying limits can occasionally impact the cluster’s utilization. For example, user-specific
limits may result in pending jobs even when many nodes are idle due to low activity.
On the other hand, these limits also enhance cluster efficiency by preventing scenarios such as a single job monopolizing all available resources, which could block other jobs from running. Without job size limits, for instance, a large job might drain the entire cluster to satisfy its resource request, a situation that is generally undesirable.
Thus, setting appropriate limits is essential to maintain fair resource usage while optimizing cluster efficiency. These limits should allow for a mix of jobs of varying sizes and types, including single-core and parallel jobs, to coexist effectively.
To implement these limits, we utilize Quality of Service (QoS). Different QoS policies are defined and applied to specific partitions in line with the established resource allocation policies. The table below outlines the
various QoS definitions applicable to the merlin7 CPU-based cluster. Here:
MaxTRESspecifies resource limits per job.MaxTRESPUspecifies resource limits per user.
| Name | MaxTRES | MaxTRESPU | Scope |
|---|---|---|---|
| normal | partition | ||
| gpu_general | gres/gpu=4 | gres/gpu=8 | user, partition |
| gpu_daily | gres/gpu=8 | gres/gpu=8 | partition |
| gpu_hourly | gres/gpu=8 | gres/gpu=8 | partition |
| gpu_gh_interactive | cpu=16,gres/gpu=1,mem=46G,node=1 | cpu=16,gres/gpu=1,mem=46G,node=1 | partition |
| gpu_a100_interactive | cpu=16,gres/gpu=1,mem=60G,node=1 | cpu=16,gres/gpu=1,mem=60G,node=1 | partition |
Where:
* normal QoS: This QoS has no limits and is typically applied to partitions that do not require user or job
restrictions.
* gpu_general QoS: This is the default QoS for gmerlin7 users. It limits the total resources available to each
user. Additionally, this QoS is applied to the [a100|gh]-general partitions, enforcing restrictions at the partition level and
overriding user-level QoS.
* gpu_daily QoS: Guarantees increased resources for the [a100|gh]-daily partitions, accommodating shorter-duration jobs
with higher resource needs.
* gpu_hourly QoS: Offers the least constraints, allowing more resources to be used for the [a100|gh]-hourly partitions,
which caters to very short-duration jobs.
* gpu_a100_interactive & gpu_gh_interactive QoS: Guarantee interactive access to GPU nodes for software compilation and
small testing.
For additional details, refer to the GPU partitions section.
Tip
Always verify QoS definitions for potential changes using the sacctmgr show qos format="Name%22,MaxTRESPU%35,MaxTRES%35" command.
GPU partitions¶
This section provides a summary of the partitions available in the gmerlin7 GPU cluster.
Key concepts:
* PriorityJobFactor: This value is added to a job’s priority (visible in the PARTITION column of the sprio -l command).
Jobs submitted to partitions with higher PriorityJobFactor values generally run sooner. However, other factors like job age
and especially fair share can also influence scheduling.
* PriorityTier: Jobs submitted to partitions with higher PriorityTier values take precedence over pending jobs in partitions
with lower PriorityTier values. Additionally, jobs from higher PriorityTier partitions can preempt running jobs in lower-tier
partitions, where applicable.
* QoS: Specifies the quality of service associated with a partition. It is used to control and restrict resource availability
for specific partitions, ensuring that resource allocation aligns with intended usage policies. Detailed explanations of the various
QoS settings can be found in the User and job limits with QoS section.
Tip
Always verify partition configurations for potential changes using the scontrol show partition command.
A100-based partitions¶
| PartitionName | DefaultTime | MaxTime | TotalNodes | PriorityJobFactor | PriorityTier | QoS | AllowAccounts |
|---|---|---|---|---|---|---|---|
| a100-general | 1-00:00:00 | 7-00:00:00 | 3 | 1 | 1 | gpu_general | merlin |
| a100-daily | 0-01:00:00 | 1-00:00:00 | 4 | 500 | 1 | gpu_daily | merlin |
| a100-hourly | 0-00:30:00 | 0-01:00:00 | 5 | 1000 | 1 | gpu_hourly | merlin |
| a100-interactive | 0-01:00:00 | 0-12:00:00 | 5 | 1 | 2 | gpu_a100_interactive | merlin |
All Merlin users are part of the merlin account, which is used as the default account when submitting jobs.
Similarly, if no partition is specified, jobs are automatically submitted to the general partition by default.
Tip
For jobs running less than one day, submit them to the a100-daily partition. For jobs running less than one hour, use the a100-hourly partition. These partitions provide higher priority and ensure quicker scheduling compared to a100-general, which has limited node availability.
GH-based partitions¶
| PartitionName | DefaultTime | MaxTime | TotalNodes | PriorityJobFactor | PriorityTier | QoS | AllowAccounts |
|---|---|---|---|---|---|---|---|
| gh-general | 1-00:00:00 | 7-00:00:00 | 5 | 1 | 1 | gpu_general | merlin |
| gh-daily | 0-01:00:00 | 1-00:00:00 | 6 | 500 | 1 | gpu_daily | merlin |
| gh-hourly | 0-00:30:00 | 0-01:00:00 | 7 | 1000 | 1 | gpu_hourly | merlin |
| gh-interactive | 0-01:00:00 | 0-12:00:00 | 7 | 1 | 2 | gpu_gh_interactive | merlin |
All Merlin users are part of the merlin account, which is used as the default account when submitting jobs.
Similarly, if no partition is specified, jobs are automatically submitted to the general partition by default.
Tip
For jobs running less than one day, submit them to the gh-daily partition. For jobs running less than one hour, use the gh-hourly partition. These partitions provide higher priority and ensure quicker scheduling compared to gh-general, which has limited node availability.