Intf-HybridMem: Page migration in hybrid memories considering cost efficiency
Introduction
Main memory capacity has become a limitation in the development of new computational systems. Cloud computing, high-scale computing systems, and big data processing are examples of the challenges that are being assigned to memory architectures. Also, there is a growing need for the development of sustainable computer systems, committed to low energy consumption. Relevant studies in the literature show that DRAM memory technology is achieving its limit of scalability [1], [2]. The main constraints of DRAM technologies are their static energy consumption and the difficulty of reducing the size of memory cells. Moreover, the charge time of capacitors increases as the size of the technology decreases [3], [4], [5]. Consequently, in DRAM technology, the grown of main memory to meet large scale computing leads to an increase in energy consumption and also in power dissipation.
Considering this scenario, new technologies have been developed as alternatives to the DRAM architecture in order to achieve high memory capacity demand. Thus, a strong trend in memory architecture is the non-volatile memories (NVMs), providing lower static power and higher density when compared to DRAM. Besides that, in some cases, NVMs can achieve an overall low energy consumption. Because of these main characteristics, NVM technologies have the potential to overcome the scalability limits related to DRAM memories.
NVM application over traditional memory architectures leads to new challenges because of the low endurance of NVM and the asymmetry time executions between read and write operations. Moreover, in these scenarios, where NVM is still under development and the simple replacement of DRAM by NVM architecture would face durability issues [6], the use of hybrid memories constitutes a new perspective to prospect solutions of such memory architecture problems. This proposal integrates the use of both approaches, DRAM and NVM, exploiting the best features of each technology [7].
One basic problem in the hybrid memory research area is the following decision-question: Which does memory module need to be select for each data writing in a request service? This decision-making problem should has to consider the uncertainties inherent to multiple characteristics of hybrid memory architectures. See, e.g., the size of memories, speed of read/write operations, endurance, and energy consumption. Moreover, and more challenging, the decision should also has to examine and identify the patterns of memory accesses, also considering the uncertainty related to all previously described attributes.
The conception of Intf-HybridMem architecture and its evaluation is presented as a proposal for migrating pages in hybrid main memory using a flexible approach based on interval-valued fuzzy logic. Fuzzy systems seek to model the uncertainties inherent of hybrid memories considering both aspects, main characteristics and distinct behavior of memory accesses, in order to support decision making problems mitigating energy consumption. And, for analyzing the performance of limit evaluations related to the Intf-HybridMem architecture, the use of an oracle migration policy is considered. Tests based on the oracle migration policy show that there is potential to better explore the main features of hybrid memories in order to achieve low energy consumption.
Section 1 shows an overview of memory technologies and hybrid memory approaches. Section 3 discuss related works. Fuzzy theory and foundations which support this work are presented in Section 2.2. Section 4 presents the architectural organization of the Intf-HybridMem. Tests, evaluation and results are discussed on Section 5, and conclusion and future works in Section 6.
Section snippets
Preliminaries
This section summarizes the studies related to main characterization of updated memory technologies and foundations underlying interval-valued inference fuzzy systems.
Data management strategies in hybrid memory studies
Researchers have proposed using NVMs to address DRAM scalability. Then, it becomes relevant to use data management in hybrid memories to deal with multiple parameters in volatile and NVMs. In this case, it is meaningful to investigate related work of data management strategies in hybrid memories.
DRAM-PCM hybrid memory architectures offer the merit of combining the advantages of DRAM and PCM and suppressing its shortcomings. Khouzani et al. [32] emphasized that PCM is a promising candidate to be
Intf-HybridMem architecture
The present work contributes to improving the data management for hybrid main memories considering cost efficiency by using a fuzzy-based approach for page migration control. Intf-HybridMem architecture considers a hybrid memory composed of two memory modules: a DRAM and an NVM module, shown in Fig. 3. This is a general approach that can admit different sizes of memories and any type of NVM. Based on NVM constraints related to endurance and energy consumption, executions dealing with
Intf-HybridMem evaluation
Tests performed used an in-house simulator to model the hybrid main memory and the Access Updater. As migration policy, we used an offline strategy, called oracle. The use of the oracle aims to define the performance limit of migration techniques in the Intf-HybridMem architecture.
As input for the tests, traces of memory access were collected by running a subset of benchmarks from Mibench [52], over GEM5 [53] and NVMain simulators [54]. We selected a subset of benchmarks for this preliminary
Conclusions and future works
In this paper, we present the Intf-HybridMem architecture, a data management proposal for Hybrid Memories based on Fuzzy Systems. For this purpose, we model a hybrid memory architecture with two modules:
- (i)
Access updater, a hardware module identifying the pages access patterns; and
- (ii)
Intf-HybridMem migration policy, a fuzzy-based software module supporting migration decision. Tests conducted aimed to identify the performance limit of the migration policies on Intf-HybridMem architecture. For this,
Conflict of interest
None declared.
Declaration of Competing Interest
The authors report no declarations of interest.
Acknowledgments
This work was supported by Brazilian Funding Agencies CAPES and CNPq; PQ Grants (310106/ 2016-8); PqG/FAPERGS 02/2017 (17/2551-0001207-0) and FAPERGS/CNPq 12/2014 – PRONEX (16/2551-0000488-9): Green-Cloud and Sustainable Computing.
References (55)
The concept of a linguistic variable and its application to approximate reasoning – I
Inform. Sci.
(1975)- et al.
Interval-valued fuzzy coimplications and related dual interval-valued conjugate functions
J. Comput. Syst. Sci.
(2014) - et al.
A cost-effective and energy-efficient architecture for die-stacked dram/nvm memory systems
2018 IEEE 37th International Performance Computing and Communications Conference (IPCCC)
(2018) - et al.
Phase-change technology and the future of main memory
IEEE Micro
(2010) - et al.
Evaluating stt-ram as an energy-efficient main memory alternative
2013 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)
(2013) - et al.
Exploring dram organizations for energy-efficient and resilient exascale memories
Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis – SC’13
(2013) - et al.
Co-architecting controllers and dram to enhance dram process scaling
The Memory Forum
(2014) - et al.
Exploiting dram restore time variations in deep sub-micron scaling
Proceedings of the 2015 Design, Automation & Test in Europe Conference & Exhibition, DATE’15, EDA Consortium, San Jose, CA, USA
(2015) - et al.
Reducing dram footprint with nvm in facebook
Proceedings of the Thirteenth EuroSys Conference, EuroSys’18
(2018) - et al.
Power and performance of read-write aware hybrid caches with non-volatile memories
2009 Design, Automation & Test in Europe Conference & Exhibition
(2009)
Overview of emerging nonvolatile memory technologies
Nanosc. Res. Lett.
Energy-and endurance-aware design of phase change memory caches
Design, Autom. & Test in Eur. Conf. & Exhibition
i2wap: Improving non-volatile cache lifetime by reducing inter-and intra-set write variations
IEEE 19th Intl. Symp. on High Perf. Comp. Arch.
Enabling a sustainable device for pervasive system architectures and applications
IEEE Intl. Electron Devices Meeting
Relaxing non-volatility for fast and energy-efficient stt-ram caches
Perf. anal. on structure of racetrack memory
23rd Asia and South Pacific Design Autom. Conf.
Approximate storage in solid-state memories
ACM Trans. Comput. Syst.
Partial-set: write speedup of pcm main memory
Energy efficient real-time task scheduling for embedded systems with hybrid main memory
J. Signal Process. Syst.
Exploring main memory design based on racetrack memory technology
Intl. Great Lakes Symp. on VLSI (GLSVLSI)
Energy-aware assignment and scheduling for hybrid main memory in embedded systems
Computing
Characterizing the overhead of software-managed hybrid main memory
2015 IEEE 23rd International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems
Function -flous, application a l’aide au diagnostic en pathologie thyroidienne (Ph.D. thesis)
Intervall-wertige mengen
Math. Nachrich.
A historical account of types of fuzzy sets and their relationships
IEEE Trans. Fuzzy Syst.
Uncertainty Data in Interval-Valued Fuzzy Set Theory - Properties, Algorithms and Applications, vol. 367 of Studies in Fuzziness and Soft Computing
Cited by (4)
Low power and high I/O efficiency hybrid memory for Internet of Things edge devices
2022, Microelectronics JournalCitation Excerpt :To improve the performance of AIoT processors, STT-MRAM can be combined with SRAM to form a hybrid cache structure with an optimized prefetch strategy [17]. Intf-HybridMem [18] proposed a hybrid memory architecture with fuzzy systems to assist with decision making. To improve the hybrid memory system, PFHA [19] proposed a hardware scheme that predicts hot pages based on periodic write and read counts.
f-HybridMem: A consensual analysis via fuzzy consensus measures and penalty functions
2022, IEEE International Conference on Fuzzy SystemsToward a Fuzzy Logic-Based Consensus Analysis in Hybrid Memory Management
2021, IEEE International Conference on Fuzzy SystemsA Proposal for Hybrid Memories Management Exploring Fuzzy-Based Page Migration Policy
2021, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)