القائمة
Trigger on SJTU HPC, 交我算, Slurm, srun, sbatch, IDEKube, hpc/sy SSH; use for cluster jobs and transfer.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Knowledge base from '交我算 HPC+AI平台用户手册' by 上海交通大学网络信息中心. Use when answering questions about HPC cluster usage, SLURM job scheduling, module environment, data transfer, GPU computing, and common troubleshooting on SJTU HPC clusters.
Log in to the SJTU HPC platform (also known as "交我算") as the user to perform job queries, submissions, cancellations, and data management. Use this skill when the user requests operations related to HPC or "交我算".
| name | sjtu-hpc |
| description | Trigger on SJTU HPC, 交我算, Slurm, srun, sbatch, IDEKube, hpc/sy SSH; use for cluster jobs and transfer. |
关键规则：
ssh hpc 或 ssh sy，不要直接用 ssh sylogin.hpc.sjtu.edu.cn（可能用错用户名）ssh hpc-data 数据节点通常不受 fail2ban 影响，可用于验证账号/证书是否正常推荐 SSH 配置 (~/.ssh/config):
Host hpc sy
HostName sylogin.hpc.sjtu.edu.cn
User YOUR_HPC_USERNAME
IdentityFile ~/.ssh/sjtu_hpc_ed25519
Host hpc-data
HostName sydata.hpc.sjtu.edu.cn
User YOUR_HPC_USERNAME
IdentityFile ~/.ssh/sjtu_hpc_ed25519
| Cluster | Login Node | Data Node |
|---|---|---|
| Siyuan-1 (思源一号) | sylogin.hpc.sjtu.edu.cn | sydata.hpc.sjtu.edu.cn |
| Pi 2.0 / AI | pilogin.hpc.sjtu.edu.cn | data.hpc.sjtu.edu.cn |
| ARM | armlogin.hpc.sjtu.edu.cn | data.hpc.sjtu.edu.cn |
| Queue | Cluster | Max Cores | Max GPU | Max Time | Use Case |
|---|---|---|---|---|---|
64c512g | Siyuan-1 | 60,000 | - | 7 days | Large CPU jobs |
a100 | Siyuan-1 | - | 92 cards | 7 days | GPU computing (A100 40GB) |
debug64c512g | Siyuan-1 | 128 | - | 60 min | CPU debugging |
debuga100 | Siyuan-1 | 64 | 28 virtual | 20 min | GPU debug (5GB vGPU, NOT real A100) |
cpu | Pi 2.0 | 24,000 | - | 7 days | CPU jobs (exclusive) |
small | Pi 2.0 | 1-20 | - | 7 days | Small jobs (shared) |
dgx2 | AI | - | 128 cards | 7 days | GPU computing (V100) |
huge | Pi 2.0 | 80 | - | 2 days | Large memory (3TB) |
Important: Login nodes prohibit running jobs and parallel compilation. For interactive work: srun -p 64c512g -n 4 --pty /bin/bash
每次操作结束后（无论正常还是异常），必须确保不留残留：
squeue -u $USER # 检查残留作业
ls $SCRATCH # 检查残留临时文件
违反后果：残留作业持续计费（浪费钱），资源滥用可能封号。
sbatch 或 srun -p 64c512g ...make -jsrun -p 64c512g -n 4 --pty /bin/bash # CPU 交互
srun -p a100 --gres=gpu:1 --pty /bin/bash # GPU 交互
| Command | Description |
|---|---|
sbatch job.slurm | Submit job |
squeue -u $USER | View your jobs |
scancel JOBID | Cancel job |
sinfo -p QUEUE | View queue status |
sacct -j JOBID | View completed job info |
module avail | List available modules |
module load NAME | Load module |
| Cluster | Proxy |
|---|---|
| Siyuan-1 | http://proxy2.pi.sjtu.edu.cn:3128 |
| Pi 2.0 | http://proxy.pi.sjtu.edu.cn:3004 |
export http_proxy=http://proxy2.pi.sjtu.edu.cn:3128 # Siyuan-1
export https_proxy=http://proxy2.pi.sjtu.edu.cn:3128
| Topic | Reference |
|---|---|
| SSH access & environment variables | references/access.md |
| Slurm job templates (CPU/GPU/Array) | references/job-templates.md |
| Python & R environment setup | references/python-r.md |
| GPU computing (A100/V100) | references/gpu.md |
| Data transfer methods | references/data.md |
| Storage architecture & quotas | references/storage.md |
| Queue specifications | references/queues.md |
| Complete Slurm commands | references/slurm-commands.md |
| HPC Studio web GUI | references/hpc-studio.md |
| Troubleshooting | references/troubleshooting.md |
| Billing & cost estimation | references/billing.md |
| Cluster monitoring dashboard | references/monitoring.md |
| Compute outsourcing workflow | references/compute-outsourcing.md |
| Parallel configuration formula | references/parallel-config.md |
| Utility scripts | references/utility-scripts.md |