HARNESS Project: Managing Heterogeneous Computing Resources for a Cloud Platform

CloudLightning: a Self-Organized Self-Managed Heterogeneous Cloud

“…The Hardware-and Network-Enhanced Software Systems for Cloud Computing (HARNESS) project [19] brings innovative and heterogeneous resources (such as FPGAs, GPUs) into cloud platforms by improving performance, security and costprofiles of cloud-hosted applications. Heterogeneous Secure Multi-level Remote Acceleration Service for Low-Power Integrated System and Devices (RAPID) [20] proposes the development of an efficient heterogeneous CPU-GPU cloud computing infrastructure, which can be used to seamlessly offload CPU-based and GPU-based (using OpenCL API) tasks of applications running on low-power devices(such as smartphones, tablets, portable/wearable devices, etc.)…”

The HARNESS Platform: A Hardware- and Network-Enhanced Software System for Cloud Computing

“…• The ConPaaS agent, which performs management actions on behalf of the Application Manager: it configures the VM where the Application Manager resides, installs code/data resources such as GPGPUs, FPGAs and XtreemFS volumes, configures access to heterogeneous resources, starts the application, and finally collects application-level feedback during execution [36]; • The Executive is a scheduling process that given a fixed set of provisioned heterogeneous compute resources, selects the most appropriate hardware accelerator for a given application task [20], [29]; • The XtreemFS client is in charge of mounting XtreemFS volumes in the VMs and making them available as regular local directories [38].…”

Quality of Service Driven Runtime Resource Allocation in Reconfigurable HPC Architectures

“…We also consider general, user-provided hardware implementations, while Hydrogen assumes a fixed repository (library) of highperformance commonly used kernels. Finally, the Maxeler platform used in this work is already harnessing the ideas of virtualization and elasticity [11]. However its runtime allocation mechanism aims to maximize DFE utilization without taking into account any QoS metrics: this results in a poor job for applications with different requirements sharing the same MPC-X.…”

“…This scenario not only increases the renting costs for the user, but is not even beneficial for the datacenter owner: to achieve high efficiency (thus reducing the datacenter management costs), all the running machines in a cluster must have a high utilization. Current solutions featuring FPGA accelerators are no different and, for this reason, research is moving in this direction [11]; however, previous approaches in the reconfigurable computing domain do not consider the Quality of Service (QoS) delivered to customers, which is of utmost importance in a utility computing paradigm. Furthermore, unlike runtime resource management solutions, static resource partitioning cannot cope with possible failures of datacenter nodes or accelerators.…”