Performance Sensitive Allocator For Hybrid Memory System

With the development of computer application and the rise of "big data",applications need more and more memory.Because of the limitation of the power consumption and manufacturing technique restrictions,Dynamic Random Access Memory(DRAM)can not meet the memory requirement of the application.Parallel hybrid memory system consists of Non-Volatile Memory(NVM)and DRAM,they can provide large capacity of NVM and lower read/write latency,lower write energy consumption,longer lifetime via DRAM.Owing to these advantages,the parallel hybrid memory system has attracted more and more attention and brought new opportunities for the development of computer architecture.To fully exploit the advantages of both NVM and DRAM,a primary goal is to properly place application data on the hybrid memories.Previous studies have focused on page migration schemes to achieve higher performance and energy efficiency.However,those schemes all rely on online page access monitoring(costly),and data migration at the page granularity is also costly in term of wasted DRAM bandwidth and inefficient use of DRAM resource.This paper presents Object-level memory Allocation and Migration(OAM)mechanisms for hybrid memory systems.OMA exploits a profiling tool to characterize objects' memory access patterns,and applies a performance/energy model to direct the application data placement on NVM or DRAM.Based on our newly-developed programming interfaces for memory allocation and migration in hybrid memory systems,application source codes can be automatically modified via static code instrumentation.When a program begins to execute,the runtime system makes a final decision on object placement according to the current DRAM usage.The experimental results show that OAM can significantly reduce the system energy-delay-product by 61%on average compared to a page-interleaving data placement scheme.It can also significantly reduce data migration traffic by 42%compared to the state-of-the-art page migration scheme CLOCK-DWF,while delivering 9%application performance improvement on average.