劉磊

劉磊主導（第一作者）的部分研究成果，以及合作（非一作）發表如下論文：

1.Memos: A Full Hierarchy Hybrid Memory Management Framework (Short Paper)

Lei Liu

2.

Lei Liu

3.

Lei Liu

4.Going Vertical in Memory Management: Handling Multiplicity by Multi-policy

Lei Liu

5.BPM/BPM+: Software-based Dynamic Memory Partitioning Mechanisms for Mitigating DRAM Bank-/Channel-level Interferences in Multicore Systems

Lei Liu

6.A Software Memory Partition Approach for Eliminating Bank-level Interference in Multicore Systems

Lei Liu

7.WiseThrottling: A New Asynchronous Task Scheduler for Mitigating I/O Bottleneck in Large-Scale Datacenter Servers

Lei Liu

8.Dynamic I/O-Aware Scheduling for Batch-Mode Applications on Chip Multiprocessor Systems of Cluster Platforms

Lei Liu

研究項目介紹

1. DRAM Bank and Channel Partitioning Mechanism (BPM/BPM+) on Real Systems

This work begins with the contention and interference issue in main memory systems, and I approach it from the Operating System angle. In existing OS, memory resources are "blindly" allocated to applications (threads), leading to memory contentions in DRAM Banks in the root. Order to solve this problem, I use a software method that is an extension of well-known Page-Coloring to eliminate/mitigate the interferences across threads in main memory. The efforts are in PACT-2012 and ACM TACO-2014.

2. "Going Vertical" in Memory Management

The "Gap" between architecture and operating system research brings challenges to computer systems. One typical issue is that the memory management mechanism in existing OS or runtime is oblivious to architecture details in memory systems, thus always leading to serious memory contentions and underutilization. In this project, the approach "Going Vertical" and "Vertical Partitioning" (having the idea of B/C/O-bits in ISCA-2014) are proposed to vertically mitigate the memory interferences in the entire memory hierarchy, as well as improve the memory utilization. Based on them, we further devise dynamic memory optimization and Curve-Vertical Partitioning approach to handle the diverse memory behaviors exhibited by the appearing "memory-diversity" workloads on multi-/many-core platforms. The efforts are published in ISCA-2014 and IEEE TC-2016 (Featured article invited).

3. Memos: Embracing Hybrid Memory Management in Modern Operating System

In this project, we try to develop a memory-centric computing model, operating system, and architecture with hybrid and heterogeneous memory systems. Our goal is to provide a new computing technology, which is not limited by "Memory Wall" in the era marked as "DRAM technology scaling is ending".

Our recent work introduces memos, a framework that can flexibly and efficiently manage main memory system with “horizontal” DRAM-NVM (or Fast-Slow) architecture. Memos has several advantages: (1) Memos integrates several typical memory policies that can benefit the NVM performance and the overall memory utilization, such as memory bank partitioning/re-balancing and vertical memory optimization approaches. (2) Powered by an OS kernel level monitoring tool, memos can obtain the memory patterns online, and then leverage them to guide the memory preference data mapping. (3) Memos can schedule appropriate cache, DRAM, NVM, and channel resources together according to memory features (or user demands), and thus achieving a higher performance. We test memos with diverse workloads, including Memcached. Current experimental results show that memos can benefit memory utilization, and contribute to system throughput and QoS.

4. SysMon

A light-weight OS level system monitoring tool suite, which is able to profile the system on-the-fly and get the memory utilization (including cache utilization, memory footprint, approximate row-buffer locality, physical page level logic re-use time, access frequency, hot/cold features and write/read patterns) without any hardware or PMU supporting. SysMon is especially useful in VM and system level research work. Sysmon is now open source on Github. The beta version is introduced in ISCA-2014 and TC-2016.

項目參與者:

ICT: Lei Liu, Mingjie Xing, Zehan Cui (2011~2013), and Chenyong Wu (2010~2014).

PITT and VMware: Yong Li (2013~2016).

Students: Hao Yang, Mengyao Xie and Hongna Geng.

資金支持:

NSFC under grant No. 61502452 (PI: Lei Liu).

Innovation research project support, SKL (PI: Lei Liu).

863 Program under grant No.2012AA010902 (PI: Xiaobing Feng).

973 Program under grant No.2011CB302504 (PI: Chengyong Wu).