Use Cases

We have planned four distinct use cases to be used for evaluating the HIGHER platform.

HIGHER will put together representative software stacks for a range of use cases of (i) accelerated data processing and analysis, for converged platforms combining Edge and Cloud processing, (ii) Infrastructure-as-a-Service, with standardised management and monitoring, (iii) Platform-as-a-Service, supporting large-scale data processing for ML inference and data analytics, and (iv) memory pool management, at the server rack level, with access control safeguards, based on current maturing CXL standards. 

(a) Use cases built on top of compute, storage and networking resources provided by server and/or racks. (b) Distributed multi-site resources, managed via Meta-OS middleware.

UC1 will demonstrate the acceleration of data processing and memory-intensive applications in CloudSuite, MLPerf benchmark collections (training and inference), and a molecular docking drug discovery application. This use case will focus on demonstrating the advantage of the HIGHER platform and its software stack on i) accelerating their performance on different cloud servers utilising ARM and RISC-V host CPUs and acceleration RISC-V-based modules, ii) cost efficiency improvement triggered by enhanced resource utilisation, and iii) energy improvements from offloading data processing close to the data source on the edge. Different stages in the use case stress different resources (i.e computation, memory, and I/O); in that respect will use designed resources of the cloud to accelerate the performance of memory-intensive and data intensive applications and explore the cloud-edge continuum by offloading processing to edge devices.   

UC2 is primarily concerned with demonstrating performance and efficient parity between mainstream x86, ARM and RISC-V-based cloud server platforms. The use case will focus on benchmarking and optimisation to achieve better performance metrics when compared to established x86-based cloud server platforms, particularly in scenarios wher the I/O operations are critical, such as storage-intensive applications. CLOUDSIGMA’s proprietary cloud computing platform will be used during the evaluation and validation phase to compare the performance results of the HIGHER platform with a fully featured commercial public cloud infrastructure.

UC3 will demonstrate a complete development and deployment environment supporting large-scale data processing for ML inference and data analytics, including utilisation of both ARM and RISC-V host systems and RISC-V accelerators. The PaaS environment developed on top of the HIGHER platform will support the entire lifecycle of applications, including build/development, test, deployment, run-time management, updates, and integration with CI/CD. 

UC4 will demonstrate the ability of applications running on the host processor to dynamically attach and transparently (to the user application) access disaggregated cache-coherent memory resources (CXL.mem) over CXL-based low-latency / high-throughput links. The use case will focus on the development of FPGA glue logic on the host ARM and RISC-V CPUs and CXL.mem memory sides that processes and packetises memory requests into CXL flits that are transmitted over the physical channel (PCIe) and vice versa. Towards evaluating throughput overheads when accessing remote CXL.mem resources, UC4 will execute well-established benchmarks (e.g. STREAM). Moreover, UC4 will also evaluate key-value store-based applications, mostly focusing on No-SQL databases using well established benchmarks, such as Memcached and Redis.