3. Computing Resources Overview¶

This chapter helps users decide which login entry, computing resource, user group, and detailed documentation they should use before submitting jobs.

Note

Login nodes are intended for login, editing files, small-scale debugging, preparing data, and submitting jobs. Production computing tasks should be submitted to HTCondor or Slurm instead of running directly on login nodes for a long time.

3.1 Choosing Computing Resources¶

Use case	Recommended resource	Description	Details
Log in, edit scripts, and check results	Login nodes	For interactive operations and job submission	SSH Login
Many independent single-core or serial jobs	HTCondor	Suitable for high-throughput workloads such as batch processing and parameter scans	HTCondor Jobs
Multi-core, MPI, long-running, or node-resource-aware jobs	Slurm CPU	Suitable for high-performance and parallel computing	Slurm Jobs
GPU computing, machine learning, GPU-accelerated simulation	Slurm GPU	GPU resources must be requested explicitly and follow group and queue policies	Slurm Jobs
Quantum-circuit GPU simulation	Quantum GPU resources	Requires membership in the `QC` user group and the quantum software environment	Quantum Computing
Experiment-specific workflows or software	Experiment resources	Experiments may have dedicated directories, software, and job rules	Supported Experiments
Off-site or collaboration resources	Remote sites	Used for corresponding collaboration sites and distributed computing workflows	Remote Sites

Login nodes are the unified entry points to the computing environment. Users usually perform the following operations on login nodes:

log in to the cluster;
edit and debug scripts;
check files and job results;
load software environments;
submit, query, and cancel jobs.

Common login entries include:

Entry	Main use
`lxlogin.ihep.ac.cn`	Main public login entry
`lhmtlogin.lhaaso.ihep.ac.cn`	LHAASO-related login entry
`hlogin.ihep.ac.cn`	HXMT-related login entry

For login methods and client configuration, see SSH Login.

3.3 Local Computing Resources¶

3.3.1 HTCondor Cluster¶

HTCondor is designed for high-throughput computing and is suitable for running many independent serial jobs. Typical use cases include:

large-scale batch data processing;
independent analysis of many input files;
parameter scans;
single-core jobs or jobs with modest resource requirements.

HTCondor jobs are usually submitted, queried, and removed with commands such as hep_sub, hep_q, and hep_rm. See HTCondor Jobs.

3.3.2 Slurm CPU Cluster¶

The Slurm CPU cluster is designed for high-performance computing tasks that need explicit CPU core, memory, wall-time, partition, or queue settings. Typical use cases include:

multi-core parallel computing;
MPI jobs;
long-running computing tasks;
jobs that must follow experiment-specific resource policies.

See Slurm Jobs.

3.3.3 Slurm GPU Cluster¶

GPU resources are suitable for programs that can benefit from GPU acceleration, such as deep learning, GPU-accelerated simulation, and some scientific computing programs. GPU jobs must request GPUs explicitly in the job script and use the appropriate partition, account, and QOS.

Before using GPUs, check:

whether the program actually supports GPU acceleration;
whether the user group has GPU resource authorization;
whether the required CUDA, driver, deep-learning framework, or experiment software is loaded;
whether the input and output directories are writable from compute nodes.

See Slurm Jobs.

3.4 Remote Site Resources¶

Some collaboration institutions provide distributed computing sites. Users can use the corresponding site resources according to their experiment or collaboration project. The remote sites currently kept in this manual are:

Remote sites usually have their own login entries, account requirements, job submission methods, and storage directories. Read the corresponding site documentation before use.

3.5 Accounts, User Groups, and Resource Authorization¶

Computing resources are usually authorized by user group or experiment group. When applying for an account or a secondary group, choose the application or resource group according to actual needs.

Common workflow:

Apply for a unified authentication account and cluster account.
Apply for the corresponding user group according to the experiment or application.
Apply for a secondary group if additional resources are needed.
After authorization, submit jobs to the corresponding cluster or queue.

For account and secondary-group application, see Account Application.

3.6 Relationship With Storage Resources¶

Whether a computing job runs reliably depends not only on CPU or GPU resources, but also on whether the data directory can be accessed properly from compute nodes.

General recommendations:

put login configuration, small files, and important personal configuration in AFS;
put temporary job files and intermediate results in work or temporary directories that compute nodes can read and write;
put experiment data in experiment-designated data directories;
use tape storage or other long-term storage for archival data;
pack large numbers of small files before transfer or long-term storage to reduce pressure on parallel file systems.

For details, see Storage and Files.

3.7 Common Entry Points¶

Task	Entry
Apply for an account	Account Application
Log in to the cluster	SSH Login
View file systems	Storage and Files
Submit HTCondor jobs	HTCondor Jobs
Submit Slurm jobs	Slurm Jobs
Use quantum computing resources	Quantum Computing
View experiment-specific instructions	Supported Experiments
FAQ	FAQ

🤖 AI 助手