BROWSE
Detail
Conference Paper
Design of Granted Memory Management for Distributed Memory Node with multi-HCAs and Large Capacity Memory
Cited - time in 
Share 
Share
- Authors
- Young-Ho Kim, Shin-Young Ahn, Eun-Ji Lim, Gyu-Il Cha
- Issue Date
- 2014-12
- Citation
- International Conference on Internet (ICONI) 2014, pp.1-6
- Language
- English
- Type
- Conference Paper
- Abstract
- In this paper, we propose a Topology-aware based granted memory management mechanism for large capacity distributed memory node. Without the consideration for the topology inter connection of system resources such as CPU, RAM, and HCA (Host Channel Adapter) s on distributed memory node which has multiple HCAs and hundreds of GB large capacity memory in memory virtualization system, it may cause the increment of latency and response time of memory service due to the access and service the memory region not adjacent memory from the operated HCA channel device. The proposed Topology-aware granted memory management is designed to memory service operations such as granted memory initialization, registration, allocation and memory transfer based on Topology map which is extracted and built from the topology interconnection and resources information such as CPU, RAM, and HCA on the memory granted node. The proposed mechanism may be minimized latency and response time of memory service of memory virtualization system, result in improvement the performance of Big-data applications such as Predictive Analytics, Content Analytics, and Bio-Informatics which is executed with data-intensive or memory-Intensive workloads.
- KSP Keywords
- Big-data, Bio-informatics, Host Channel Adapter(HCA), Large capacity, Management mechanism, Memory Transfer, Memory initialization, Memory intensive, Memory management, Memory virtualization, Topology map