- Briefing
- Create NSCC account
- Connect to NSCC server
- Upload a project
- Train a deep learning model
- Done.
- A Sample Here !!!
Briefing
NSCC (National Supercomputing Centre) Singapore is a free super computer resource for the students/staffs from NUS, NTU, SMU and SUTD.
The steps below more for NUS.
Create NSCC account
- go to https://www.nscc.sg/
- click “FOR NEW USERS ENROLLMENT” -> login -> select NUS -> login to school account -> Done ! You can see the NSCC User Portal.
- Setup your SSH Key or password (alternative) in the NSCC User Portal.
- (if you choose SSH Key, for Mac users, you just copy the content and store in a file named id_rsa, throw in the ~/.ssh folder; for Win users, store as a file & configure in the putty.)
Connect to NSCC server
If you are not in campus, you need use VPN!
For NUS student, please refer to this PDF. From https://nusit.nus.edu.sg/eguides/.
Connect with SSH command (Mac)
Check this NSCC New User Guide
ssh e0015130@nus.nscc.sg
You will see your remaining computing time… For new user is 100000 hrs.
Connect with Putty (Win)
Download Putty, https://www.putty.org/.
Connect to nus.nscc.sg with username / password OR ssh key.
Login! Check around
module available
git --version
Upload a project
- You can connect to the server use FTP tools as Cyberduck/Mac or FileZilla/Win, then upload your project.
- You can download project from internet.
- Use git command clone project. (As the sample here)
Train a deep learning model
Prepare a PBS script ask for the computing resources
vi submit.pbs
#!/bin/bash
#PBS -k oe
#PBS -q gpu
#PBS -j oe
#PBS -l select=1:ncpus=24:mem=80GB
#PBS -l walltime=24:00:00
#PBS -N resnet_bde
#PBS -P Personal
cd ${PBS_O_WORKDIR}
module load python/2.7.12
module load tensorflow/1.0+keras
python custom_resNet.py > output_log.txt
Submit job use PBS commands
qsub submit.pbs
You can use below PBS commands to add job, check job status or del job.
[e0015130@nus04 v7]$ qsub submit.pbs
6841177.wlm01
[e0015130@nus04 v7]$ qstat
Job id Name User Time Use S Queue
---------------- ---------------- ---------------- -------- - -----
6841177.wlm01 resnet_spp e0015130 0 Q gpunormal
[e0015130@nus04 v7]$ qdel 6841177.wlm01
[e0015130@nus04 v7]$ qstat
Done.
Pros
- FREE for student ! Although got total computing time limit as 100000.00 hrs, not a big issue.
Cons
- Seems only can run 1 job at 1 time.
- Can not upgrade the Keras version, can use some newly added modle.
- Some time hard to get the job run due to a long queue.
- Reminder do housekeeping! Any files more than 30 days old may be removed without notice!
A Sample Here !!!
After login to NSCC:
(Sorry, encounter some problem clone the project from github directly, you might need download the project first then upload via FTP)
git clone https://github.com/shenshutao/cnn_model.git
cd cnn_model/
qsub submit.pbs
qstat