March 20, 2025 | 3:10 - 4:00 PM

Location: Kemper 1003

Speaker:  Rafael Alejandro Diaz Cruz

Topic: QMCPack with gem5

March 13, 2025 | 3:10 - 4:00 PM

Location: Kemper 1003

Speaker: Nishant Acharya

Abstract: Mission Driven organizations that provide high impact local services like schools, libraries, hospitals and others, require access to an adequate internet connection as they often supplement or replace local connectivity. Past events have shown that these institutions, which we collectively call Anchor Institutions (AINs), have inadequate internet connections. As such, there is a need to monitor an AIN's connection reliability and performance, but to the best of our knowledge there is no way to accurately determine an AIN's network, which is needed to measure the connection reliability and performance. In this study, we aim to identify an AIN's network without having access to any machine in that network, in order to generalize the identification framework. In order to achieve this goal, we attempt to merge various data sources (WHOIS data, Commercial geolocation services, Broadband map) and techniques (Active measurements, Port Scanning, Reverse Geolocation) to come up with a confidence score. One of the key challenges of this study is to combine multiple, inherently unreliable, data sources and create a scalable framework to achieve the goal of identifying the network space of an AIN.

Bio: Nishant is a 4th year PhD student working with Dr. Alexander Gamero-Garrido. His research focuses on the intersection of computer networking systems, network infrastructure analysis and privacy, and aims to understand questions about the availability, fragility and efficiency of the service to networking systems.

March 6, 2025 | 3:10 - 4:00 PM

Location: Kemper 1003

Speaker: Zhantong Qiu

Topic: Nugget: Cross-platform Realistic Program Snippets

Abstract: Evaluating application performance on new hardware is crucial to computer architecture research, but it can be extremely time-consuming, especially for non-existent hardware, with cycle-level software simulation potentially causing over a 100,000X slowdown. One approach to reducing the size of applications is to sample the application to automatically find representative intervals of different application phases. However, prior sampling techniques have limitations: finding representative intervals is expensive, validating representative intervals is infeasible, and samples do not generalize across architectures or to full system evaluation. We present a new methodology for generating cross-platform program snippets. We call these snippets nuggets. We implement the Nugget methodology through compile-time program modifications, specifically leveraging LLVM Intermediate Representative (LLVM IR) and LLVM compiler passes. We overcome the limitations of prior sampling approaches by enabling low overhead and fast analysis using hardware, producing architecture- and micro-architecture-independent program snippets that can be executed on both real hardware and simulator to validate the chosen samples on real hardware. We demonstrate that the selected nuggets can be generalized across different systems and can represent multi-threaded intervals for full-system evaluation. Testing on hardware, with NPB input size C and 8-thread, we show that the average performance prediction error in runtime of the same set of generated nuggets on x86 and ARM is 5.8%. The average performance prediction error in speedup between the x86 and ARM machines is 4.86%. With the Nugget methodology, researchers will be able to find relevant program snippets in full-scale applications enabling evaluation of applications in simulation which are infeasible today.

February 27, 2025 | 3:10 - 4:00 PM

Location: Kemper 1131

February 20, 2025 | 3:10 - 4:00 PM

Location: Kemper 1003

February 6, 2025 | 3:10 - 4:00 PM

Location: Kemper 1003

Topic: Deterministic Networking Frameworks for Latency Critical Large Scientific Data Transfers

Speaker: Vijeth K L

December 11, 2024 | 2:00 - 3:00 PM

Location:  Physical and Data Sciences Building Room 1025

Topic: Emerging Challenges for Future Memory Systems

Speaker: Steven Woo

Abstract: Across all markets, computing systems are being driven to achieve higher performance in the face of rising power, cost, and reliability challenges. These systems are increasingly reliant on their DRAM memory, one of the semiconductor industry’s foundational components, to achieve system performance and power-efficiency targets. As the industry looks towards the future, increasing DRAM and memory systems performance is becoming more difficult as new reliability challenges emerge. In this talk I’ll describe some of our research work on RowHammer for server memory systems, and some of the key challenges future DRAMs and memory systems face in AI systems as data rates increase.

Bio:  Steven Woo is a Fellow and Distinguished Inventor at Rambus Inc., working on technology and business development efforts across the company. He currently leads research within Rambus Labs on advanced memory systems for accelerators and computing infrastructure, and manages a team of senior architects. Since joining Rambus, Steve has worked in various roles leading architecture, technology, and performance analysis efforts, and in marketing and product planning roles leading strategy and customer programs. Steve received his PhD and MS degrees in Electrical Engineering from Stanford University, and Master of Engineering and BS Engineering degrees from Harvey Mudd College.

December 5, 2024 | 3:10 - 4:00 PM

Location: Physical and Data Sciences Building Room 1025

Topic: Cloud System Security: From Cloud Orchestration to Hardware Micro-Architectural Attacks

Speaker: Chongzhou Fang

November 21, 2024 | 3:10 - 4:00 PM

Location: Kemper Hall room 1131 and 1127

October 24, 2024 | 3:10 - 4:00 PM

Location: Kemper Hall room 1131 and 1127

Topic: GlucOS: Security, correctness, and simplicity for automated insulin delivery

Speaker: Hari Venugopalan

Materials: Paper and Slide

October 10, 2024 | 3:10 - 4:00 PM

Location: Kemper Hall room 1131 and 1127

Topic: For the first Computer System Research Seminar, we will be presenting our cutting-edge research and projects.