General Settings

All applications listed in the UCloud apps catalogue are packaged in Docker container images and deployed on the YouGene HPC cluster, hosted at Syddansk Universitet.

The YouGene supercomputer consists of 87 compute nodes with a total of 2784 CPU cores and a theoretical maximum performance of 1.05 Tflops per CPU. Each node has either 384 GB or 768 GB of RAM.

By default each application runs on a single node of the cluster, unless multi-node deployment is enabled. The general app configuration settings are summarized below.

Job lifetime

The user should always estimate the time necessary to complete the run before submitting a job. A reliable estimate of the program execution time is important to ensure fast job scheduling and completion.

Note

There is no upper limit to the job lifetime. The time allocation can also be extended at runtime.

Machine type

Before submitting a job, the user must select a machine type. The latter depends on the products available in the active workspace.

Note

Selecting a machine with a large number of resources may result in a longer job scheduling.

UCloud standard nodes

There are seven machine types:

  1. u1-standard-1:

    1 vCPU and 5 GB of memory

  2. u1-standard-2:

    2 vCPUs and 11 GB of memory

  3. u1-standard-4:

    4 vCPUs and 23 GB of memory

  4. u1-standard-8:

    8 vCPUs and 47 GB of memory

  5. u1-standard-16:

    16 vCPUs and 94 GB of memory

  6. u1-standard-32:

    32 vCPUs and 188 GB of memory

  7. u1-standard-64:

    64 vCPUs and 376 GB of memory

UCloud fat nodes

There are seven machine types:

  1. u1-fat-1:

    1 vCPU and 10 GB of memory

  2. u1-fat-2:

    2 vCPU and 22 GB of memory

  3. u1-fat-4:

    4 vCPU and 47 GB of memory

  4. u1-fat-8:

    8 vCPU and 94 GB of memory

  5. u1-fat-16:

    16 vCPU and 188 GB of memory

  6. u1-fat-32:

    32 vCPU and 376 GB of memory

  7. u1-fat-64:

    64 vCPU and 754 GB of memory

UCloud GPU nodes

There are twelve machine types:

  1. u1-gpu-1:

    1 NVIDIA V100 GPU, 16 vCPUs, and 44 GB of memory

  2. u1-gpu-2:

    2 NVIDIA V100 GPUs, 32 vCPUs, and 88 GB of memory

  3. u1-gpu-3:

    3 NVIDIA V100 GPUs, 48 vCPUs, and 132 GB of memory

  4. u1-gpu-4:

    4 NVIDIA V100 GPUs, 63 vCPUs, and 180 GB of memory

  5. u2-gpu-1:

    1 NVIDIA A100 GPU, 12 vCPUs, and 252 GB of memory

  6. u2-gpu-2:

    2 NVIDIA A100 GPU, 24 vCPUs, and 504 GB of memory

  7. u2-gpu-3:

    3 NVIDIA A100 GPU, 36 vCPUs, and 756 GB of memory

  8. u2-gpu-4:

    4 NVIDIA A100 GPU, 48 vCPUs, and 1008 GB of memory

  9. u2-gpu-5:

    5 NVIDIA A100 GPU, 60 vCPUs, and 1260 GB of memory

  10. u2-gpu-6:

    6 NVIDIA A100 GPU, 72 vCPUs, and 1512 GB of memory

  11. u2-gpu-7:

    7 NVIDIA A100 GPU, 84 vCPUs, and 1764 GB of memory

  12. u2-gpu-8:

    8 NVIDIA A100 GPU, 96 vCPUs, and 2016 GB of memory

AAU general nodes

There are four machine types:

  1. uc-general-small:

    4 vCPUs and 16 GB of memory

  2. uc-general-medium:

    8 vCPUs and 32 GB of memory

  3. uc-general-large:

    16 vCPUs and 64 GB of memory

  4. uc-general-xlarge:

    64 vCPUs and 256 GB of memory

AAU GPU nodes

There is only one machine type available:

  1. uc-t4-1:

    1 GPU, 10 vCPUs and 40 GB of memory

Note

uc-general and uc-t4 are virtual machines deployed on the AAU OpenStack system.

Import data

A folder can be attached as a data volume inside the application container using the button


in the front-end application page. Data volumes are mounted within the /work directory inside the application container. This also corresponds to the default working tree on UCloud.

Data volumes can also be mounted in multiple apps running simultaneously.

Important

Only files and folders located in the default working tree are saved after job completion.

Multi-node deployment

Computation distributed among multiple nodes of the cluster is enabled only for few supported applications. See the Spark Cluster app for a practical use case.

Connect to other jobs

This option is used when it is necessary to use services from other jobs as networking and shared application file systems. By clicking on


the user is able to select the ID of a running job and set a hostname parameter, which is used to assign an IP address to the node where the selected job is executed.

Attach public IP addresses

This option is used to attach a static IP address to an app deployed on UCloud. In this way it is possible to access the app via an external client. Public IPs may be used to deploy server applications (see, e.g., Rsync Server, MariaDB Server, PostgreSQL Server).

To create a new IP address, click on

and select a provider and a product, as shown below:

../_images/allocate-ip-addresses.png

The IP address is unique: It is not possible to select the same IP for multiple job sessions running at the same time.

Once the IP address is allocated, the user can configure the protocol (TCP/UDP) and the corresponding port number. Project admins can also restrict the usage of specific IPs to a selected group of collaborators.

Important

By enabling this setting, anyone with the IP can contact the application. Actions must be taken to ensure that the application is adequately protected.