Deployment of AWS cluster with custom odyhpc.com AMIs
(v – Using the master instance for high-performance display)

One of the advantages of using AWS for HPC computations is the availability of more flexible & cheaper options for storage and post-processing purposes than traditional on-premises infrastructure. Many HPC apps require high-performance visualization capabilities as part of their post-processing tasks. In addition to a wide selection of instances with acceleration capabilities, AWS offers NICE-DCV (https://aws.amazon.com/hpc/dcv/) that is a high-performance remote display protocol useful for HPC and other graphically intensive workloads. A single instance can host NICE-DCV but the master node of a cluster can also serve as the hosting hardware. To accomplish this task, the cluster configuration script must be modified accordingly. The ‘configuration_script_details’ file in the CLUSTERs subdirectory provides the details for examples (e.g. DCV_1) where the master instance also hosts NICE-DCV.
Hosting NICE-DCV is usually much more computationally expensive than most other tasks handled by the master instance. It is therefore recommended to select an instance larger than when no hosting NICE-DCV. It is impossible to provide specific rules for this selection as there are too many variables that impact performance. Users will have to try and test different choices depending on their goals. However, we can provide some general recommendations to begin the process. When using x86_64 architecture, the choice will depend on the intended graphic-processing use. If nothing or very slight, an instance with larger memory (e.g. r5 series) should suffice. If the purpose is graphic intensive, we recommend trying P2 (Nvidia K-80) and P3 (Nvidia V-100) instances. For extreme graphic processing capabilities, P4 instances with Nvidia A-100 are also available, but users should be aware of the price tag as they are some of the most expensive instances offered by AWS.

Close Menu