Sunfish Use Cases

Sunfish provides a centralized digital representation of a compute system with the ability to gather and report run-time conditions and associations, using Redfish and Swordfish.  These digital avatars provide easier management of resource associations and network link-ups.  The OpenFabrics Alliance has created the OpenFabrics Enterprise Distribution stack (OFED) for Remote Direct Memory Access (RDMA) and kernel bypass  The CXL Consortium has been working to provide new computer architectures with the specifications to provide architectures to provide memory sharing, memory pooling, resource sharing, and resource aggregation.  Taken together, these new technologies provide the ability to implement composable distributed HPC and cloud computing systems.

New Composable systems have the potential to provide advantages to current non-Composable systems.  In current HPC systems, each compute node contains a fixed quantity of CPUs, memory resources, and accelerator cards.  Likewise, current network links are designed to provide pre-ordained tasks, whether the task is run-time data exchanges, IO, or management.  Resources cannot be shared or exchanged due to the fact that current computer architectures and environments don't provide the resource recognition and infrastructure to manage resources.  This leads to unfortunate limitations and built-in computational inefficiencies.  The figure below shows an example of a current HPC system that has unshared resources and a Composable HPC system that mitigates wasted resources by allowing management and sharing of the resources.

Another aspect of Composable systems is both security and multi-tenancy.  For instance, resources and 'views' can be provided to various sets of users based upon their security levels and their level of responsibility.

Finally, as demonstrated in the Use Cases, Composability provides system architectures that are better able to compute and find relationships hidden within data and better scale both data storage and system resources.