BROWSE
Detail
Conference Paper
End Performance SLA Support for Disaggregated Memory
Cited - time in 
Share 
Share
- Authors
- Kwangwon Koh, Kangho Kim, Changdae Kim, Jaehyuk Huh
- Issue Date
- 2019-04
- Citation
- Workshop on Resource Disaggregation (WORD) 2019, pp.1-5
- Language
- English
- Type
- Conference Paper
- Abstract
- This paper proposes a new service-level agreement (SLA) support for disaggregated memory, which consists of fast direct DRAM and slow indirect memory in remote machines or local non-volatile memory. Based on a new performance prediction model, the SLA support provides the bounded performance as compared to the ideal configuration only with the direct memory. For the SLA fulfillment, we propose an efficient performance predictor using indirect memory traces, and a memory management system to adjust the local memory capacity as necessary. We implement the proposed SLA support in a disaggregated memory system integrated to Linux KVM, and show its feasibility with real machine experiments. By providing the performance bound, the proposed SLA support facilitates the adoption of disaggregated memory in cloud computing, where the performance consistency is critical.
- KSP Keywords
- Cloud Computing, Local memory, Management system, Memory System, Memory management, Non-Volatile Memory(NVM), Performance Bounds, Performance consistency, Performance prediction model, Service-level Agreement(SLA), System Integrated