Skip to content

Commit

Permalink
🍱 Add CSE shared server doc
Browse files Browse the repository at this point in the history
  • Loading branch information
ZihaoZhou committed Oct 3, 2024
1 parent d6330f0 commit a5ac17b
Show file tree
Hide file tree
Showing 3 changed files with 80 additions and 5 deletions.
14 changes: 13 additions & 1 deletion docs/.vitepress/config.mts
Original file line number Diff line number Diff line change
Expand Up @@ -44,11 +44,23 @@ export default defineConfig({
},

nav: [
{ text: 'Guide', link: '/guide/', activeMatch: '/guide/' },
{ text: 'Roselab Guide', link: '/guide/', activeMatch: '/guide/' },
{ text: 'CSE Public Server Guide', link: '/shared/', activeMatch: '/shared/' },
{ text: 'Config', link: '/config/', activeMatch: '/config/' }
],

sidebar: {
'/shared/': [
{
text: 'Shared',
items: [
{
text: 'Shared Servers',
link: '/shared/',
},
],
}
],
'/guide/': [
{
text: 'Guide',
Expand Down
8 changes: 4 additions & 4 deletions docs/guide/index.md
Original file line number Diff line number Diff line change
Expand Up @@ -205,6 +205,10 @@ To see the file system, run `df -H` . You would see
* a 5 TB private data HDD mounted under `/data` that is only accessible to you, and
* a 5 TB public data HDD mounted under `/public` that is accessible to everyone.

::: tip Note
The system SSD is faster but has limited space and is local to the machine. The HDD is synchronized between machines and is suitable for large datasets. When you modify files in `/data` in one container, the changes will be reflected in your containers on all other machines. Use the HDD for larger or archived project data.
:::

```bash
$ df -H
Filesystem Size Used Avail Use% Mounted on
Expand All @@ -216,10 +220,6 @@ data/public 5.0T 263k 5.0T 1% /public

It is recommended to use soft links to access your data files on the `/data` HDD. For example, instead of downloading your data files to `/data/project1/sample...pt` and hard-coding their absolute paths, you can create a soft link under the code folder using the `ln -s /data/project1/ /home/ubuntu/project1/data/` command. Then, you can refer to the data files as if they and the code are in the same project structure.

::: tip
For datasets under 200 GB, use the faster system SSD. Use HDD for larger or archived project data.
:::

### 5. Verify Web Application Access

Refer to the documentation for [Seafile](https://help.seafile.com/), [HedgeDoc](https://docs.hedgedoc.org/), and [MinIO](https://min.io/docs/minio/linux/index.html) to access these services. For [Jupyter Lab](./jupyter) or [Remote Desktop](./rdp), check the respective pages for login instructions.
Expand Down
63 changes: 63 additions & 0 deletions docs/shared/index.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,63 @@
# Getting Started with CSE Shared Servers

## Overview

After the containerization migration, the CSE shared servers are now under the management of the RoseLab. The CSE shared servers (north, south, east, and west) also provide a containerized environment for research and development. Each user has their own isolated container with dedicated resources and sudo access.

## Key Differences from RoseLab

1. **Fixed Server Assignment**: Unlike RoseLab, you cannot move between machines. Your container is fixed to the assigned server. Your data directory is `/data/<username>` instead of `/data` (due to the backward compatibility with the previous setup). Your data directory is not synced between servers.

2. **No Common Utilities**: The shared servers do not have access to the common utilities available in RoseLab for managing containers or port mappings.

3. **Limited Software**: The shared servers come with a basic setup. Additional software installations are the responsibility of the user.

4. **Shared Conda**: A shared Conda installation is mounted by default to reduce storage usage. However, you are free to install your own Conda environment. You can disable the default shared Conda activation by removing the corresponding line from your `.bashrc` file.

5. **No Hosted Services**: The shared servers do not host any services (e.g., MiniO, WandB). Users are responsible for setting up their own services.

## Accessing Your Container

### SSH Access

To access your container via SSH, following the same steps as in RoseLab:

```bash
ssh ubuntu@[server].ucsd.edu -p [your-ssh-port] -i path/to/your/private-key
```

The credential email is similar to the RoseLab email, with the private key attached but without any credentials (since there is no service account). You are expected to set up your own password by running `sudo passwd ubuntu` after logging in.

## Resource Allocation

Each user container is allocated:
- 10 CPU cores
- 128 GB RAM
- 128 GB root drive
- Unlimited data drive (mounted at `/data/<username>`)

Check your allocated resources using similar commands as in RoseLab.

## File System

- Your home directory is located at `/home/ubuntu`
- Your data drive is mounted at `/data/<username>`
- Shared public data is accessible at `/data/public`

## Best Practices

1. **Data Storage**: Store large datasets and project files in your `/data/<username>` directory to preserve space on the root drive.

2. **File Sharing**: Use the `/data/public` folder for sharing files with other users.

3. **Port Usage**: Be mindful of the port mappings when setting up services. Use the provided host ports for external access.

4. **Resource Management**: Monitor your resource usage to ensure you stay within the allocated limits.

::: tip
If the data drive is full, you (or your PI) would need to purchase your own additional storage, since Roselab admins cannot delete users' files on their behalf. Roselab admins can only provide guidance on mounting your storage and mapping it to your container. Contact the RoseLab admins for more information.
:::

## Support

Roselab admins provide only limited support for the shared servers. Remember to keep your access credentials secure and do not share them with others.

0 comments on commit a5ac17b

Please sign in to comment.