Algorithm-Directed Data Placement in Explicitly Managed Non-Volatile Memory

Abstract

The emergence of many non-volatile memory (NVM) techniques is poised to revolutionize main memory systems because of the relatively high capacity and low lifetime power consumption of NVM. However, to avoid the typical limitation of NVM as the main memory, NVM is usually combined with DRAM to form a hybrid NVM/DRAM system to gain the benefits of each. However, this integrated memory system raises a question on how to manage data placement and movement across NVM and DRAM, which is critical for maximizing the benefits of this integration. The existing solutions have several limitations, which obstruct adoption of these solutions in the high performance computing (HPC) domain. In particular, they cannot take advantage of application semantics, thus losing critical optimization opportunities and demanding extensive hardware extensions; they implement persistent semantics for resilience purpose while suffering large performance and energy overhead. In this paper, we reexamine the current hybrid memory designs from the HPC perspective, and aim to leverage the knowledge of numerical algorithms to direct data placement. With explicit algorithm management and limited hardware support, we optimize data movement between NVM and DRAM, improve data locality, and implement a relaxed memory persistency scheme in NVM. Our work demonstrates significant benefits of integrating algorithm knowledge into the hybrid memory design to achieve multi-dimensional optimization (performance, energy, and resilience) in HPC.

Intelligent Data Placement on Discrete GPU Nodes with Unified Memory

“…HMS: Work on data placement in HMS have either focused on NVM+DRAM [17,26,34,[52][53][54] or DRAM+HBM systems [41,42,50]. Dulloor et al develop a placement policy where data structures are placed in DRAM based on access patterns on an emulated DRAM+NVM platform [17].…”

Compiler aided checkpointing using crash-consistent data structures in NVMM systems

“…We use LU decomposition as an example. LU decomposition factors a matrix A as the product of a lower triangular matrix L and an upper triangular matrix U [56]. The execution of LU decomposition program typically consists of a sequence of iterations.…”

Runtime Data Management on Non-Volatile Memory-based Heterogeneous Memory for Task-Parallel Programs

“…Wang et al [24] rely on static analysis and advanced memory controller to monitor memory access pa erns to determine data placement on GPU. Wu et al [26] leverage the knowledge of numerical algorithms to direct data placement. ey introduce hardware modi cations to support massive data migration and performance optimization.…”

“…Our initial performance evaluation with HPC workloads (Section 2) shows that there is 1.09x-8.4x slowdown on NVM-based systems, depending on bandwidth and latency features of NVM. Because of the limitation of NVM, NVM is o en paired with a small fraction of DRAM to form a heterogeneous memory system (HMS) [8,10,13,19,24,26]. By selectively placing frequently accessed data in the small amount of DRAM available in HMS, we are able to exploit the cost and scaling bene ts of NVM while minimizing the limitation of NVM with DRAM.…”

“…In HMS, we assume that DRAM shares the same physical address space as NVM (but with di erent addresses) and DRAM memory allocation can be managed at the user level. is assumption has been widely used in the existing work [8,10,13,19,26].…”