We’d like to wish a very warm congratulations to Christina Giannoula on her successful PhD defense. Christina defended her PhD on September 27, 2022 at her home institution, the National Technical University of Athens (NTUA), to a 7-member committee, including her co-advisor, Onur Mutlu.
During her PhD, Christina Giannoula was a visiting researcher with SAFARI in 2019, and made significant contributions to collaborative works with our group. Her contributions focused on Near-Data-Processing Architectures, Architectural Support for Synchronization, Hardware/Software Cooperation. She has continued to have strong collaborations and contributions to our group’s work on these topics.
We recently interviewed Christina about her PhD work and her future plans. Here’s what she had to say:
Can you tell us about the significance of your recent PhD papers, in collaboration with SAFARI?
Christina: In the last years of my PhD, I led two research papers in collaboration with the SAFARI Research Group: a) SynCron presented at HPCA 2021, and b) SparseP presented at Sigmetrics 2022. Both research works are particularly important and impactful, since they enable the widespread adoption and use of the emerging Near-Data-Processing (NDP) (or Processing-In-Memory (PIM)) architectures for the challenging workload area of irregular applications including graph processing, data analytics, pointer-chasing and sparse linear algebra kernels.
SynCron is the first low-cost and practical synchronization mechanism for NDP systems. The emergence of multi-core architectures created the need for synchronization among parallel threads to achieve high system performance in parallel applications. SynCron is the first research work to explore the synchronization problem in NDP architecture and makes three key contributions. First, it classifies the most common synchronization approaches in two types, and provides insights on which type of synchronization approach fits best in which architecture (processor-centric CPUs/GPUs and memory-centric NDPs) based on the architectural characteristics. Second, it provides a literature overview of the most important synchronization mechanisms proposed in the past, including the seminal papers of QLDB [Goodman et al., ASPLOS 1989], CM5 [Leiserson et al., SPAA 1992], Tera [Alverson et al., ICS 1990], HEP [Jordan, ISCA 1982], NYU Ultracomputer [Gottlieb et al., ISCA 1982], CRAY T3E [Scott, ASPLOS 1996] and SGI Origin [Laudon, ISCA 1997], and examines their applicability on NDP architectures. Third, SynCron proposes a novel end-to-end synchronization solution for the new/unconventional NDP architectures, which significantly differ from commercial processor-centric architectures (CPUs, GPUs), since they have a very shallow cache hierarchy and do not typically support cache coherence protocols. In my opinion, the reader can greatly benefit from reading SynCron, since he/she can comprehensively understand the synchronization problem in modern computing systems, learn the architectural differences between the processor-centric and memory-centric in the synchronization context, and get an overview of the state-of-the-art synchronization mechanisms proposed in the past.
SparseP is the first effective software library for the most important irregular computational kernel, the Sparse Matrix Vector Multiplication (SpMV) kernel, tailored for real PIM architectures. The SpMV kernel ranks among the most thoroughly studied irregular scientific computation kernels, since it lies at the heart of a large variety of applications (e.g., economic modeling, signal processing, document/information retrieval, large-scale simulations, medical imaging), and often constitutes their key performance bottleneck. Even though researchers have spent considerable effort to optimize the SpMV in modern processor-centric systems (CPUs, GPUs), this kernel still performs very poorly, achieving typically less than 10% of the machine’s peak performance. To this end, SparseP makes three key contributions. First, it comprehensively describes the SpMV execution on memory-centric near-bank PIM systems with thousands of PIM cores. Second, it proposes new SpMV algorithms tailored for PIM systems, and performs a rigorous characterization study to demonstrate which algorithm fits best on which sparse matrix with diverse sparsity patterns. This key contribution enables an impactful research direction of designing SpMV auto-tuners for real PIM systems to provide the best performance for all various sparse matrices. Third, SparseP provides programming recommendations for software designers, and suggestions and hints for hardware and system designers of future PIM systems. Our paper on SparseP helps researchers to effectively understand the key challenges of the execution of irregular computational kernels on real PIM systems, learn performance optimization techniques for PIM software, and understand architectural/system characteristics as well as limitations of PIM systems.
Finally, during my PhD I also contributed as a co-author to three very impactful research contributions from the SAFARI Research Group. 1) SMASH, our proposed hardware-software cooperative mechanism for sparse linear algebra computational kernels presented in MICRO 2019, paves the way to designing efficient hardware-software co-designed solutions for the important irregular applications. 2) PrIM, our extensive characterization study on the state-of-the-art real PIM system published in IEEE Access 2022, enables significant system performance improvements in multiple aspects of PIM hardware and software. 3) NATSA, our NDP accelerator for time series analysis presented in ICCD 2020, inspires future application-specific and NDP-based hardware designs for emerging irregular applications.
What do you feel was your most significant impact with your PhD work?
Christina: My doctoral thesis bridges the gap between processor-centric CPU systems and memory-centric NDP/PIM systems in the critically-important area of irregular applications. The most significant contributions are that my PhD thesis identifies (i) the challenges and key characteristics of irregular applications which highly affect their performance in multithreaded executions, and (ii) the architectural and software-level differences between processor-centric and memory-centric architectures in the context of irregular applications, and proposes novel algorithmic designs as well as practical hardware mechanisms tailored for emerging irregular applications. I hope that the ideas, methods and techniques presented in my dissertation inspire future work and new research studies in co-designing software algorithms with cutting-edge computing platforms to accelerate the execution of emerging irregular applications.
At this point, I would like to mention that my interaction and close collaboration with Prof. Onur Mutlu and the SAFARI Research Group constitute the cornerstone for my PhD research achievements. I wholeheartedly thank my advisor Prof. Onur Mutlu for the generous guidance, continuous support, valuable feedback and amazing opportunities he provided. His motivation for top-notch research, high-quality teaching and his passion for excellence have significantly shaped my research mindset, and will continue to be a huge source of inspiration for me. I would like to thank the SAFARI Research Group members for creating a rich, stimulating and highly motivating research environment during my PhD. As I mentioned in a previous interview, during my visit with the SAFARI Research Group, I realized that impactful and innovative research requires close collaboration among the team members. After my visit with SAFARI, I tightly embraced and integrated this key characteristic in my research personality.
What will be your next steps, moving forward?
Christina: During my PhD studies, I really enjoyed my time spent on acquiring enormous amounts of scientific knowledge, conducting high quality research, participating in teaching activities and interacting with colleagues and students. Thus, moving forward, I will continue working in academia and performing research in cutting-edge technologies. In winter 2023, I am happy to be joining the University of Toronto as a Postdoctoral researcher, working with Prof. Gennady Pekhimenko and his research group. I am also very happy to continue working with Prof. Onur Mutlu and the SAFARI Research Group as an affiliated senior researcher. My next research directions will focus on accelerating the execution of machine learning and deep neural network workloads in modern computing platforms, as well as leveraging machine learning techniques to improve multiple aspects of modern computing systems. I am very excited to conduct research in the intersection of computer systems and machine learning, and I look forward to my PostDoc scientific journey and collaborating with new scientists on the other side of the world!
Read more:
Christina also wrote about her experiences with our group in an interview in our January 2021 Newsletter.
Christina’s website: https://cgiannoula.github.io/
Christina Giannoula, (defended 27 September 2022)
Thesis title: “Accelerating Irregular Applications via Efficient Synchronization and Data Access Techniques“, PhD Thesis, National Technical University of Athens
[Slides (pptx) (pdf)]
[Thesis arXiv (pdf)]
[SAFARI Live Seminar Video (with Q&A); SAFARI Live Seminar Slides (pptx) (pdf)]
Core PhD Contributions with SAFARI:
Christina Giannoula, Ivan Fernandez, Juan Gomez-Luna, Nectarios Koziris, Georgios Goumas, and Onur Mutlu,
“SparseP: Towards Efficient Sparse Matrix Vector Multiplication on Real Processing-In-Memory Architectures”
Proceedings of the ACM International Conference on Measurement and Modeling of Computer Systems (SIGMETRICS), Mumbai, India, June 2022.
[Extended arXiv Version]
[Abstract]
[Slides (pptx) (pdf)]
[Long Talk Slides (pptx) (pdf)]
[SparseP Source Code]
[Talk Video (16 minutes)]
[Long Talk Video (55 minutes)]
Christina Giannoula, Nandita Vijaykumar, Nikela Papadopoulou, Vasileios Karakostas, Ivan Fernandez, Juan Gómez-Luna, Lois Orosa, Nectarios Koziris, Georgios Goumas, and Onur Mutlu,
“SynCron: Efficient Synchronization Support for Near-Data-Processing Architectures”
Proceedings of the 27th International Symposium on High-Performance Computer Architecture (HPCA), Virtual, February-March 2021.
[Slides (pptx) (pdf)]
[Short Talk Slides (pptx) (pdf)]
[Talk Video (21 minutes)]
[Short Talk Video (7 minutes)]
[SAFARI Live Seminar Video (1 hr 29 mins)]
Konstantinos Kanellopoulos, Nandita Vijaykumar, Christina Giannoula, Roknoddin Azizi, Skanda Koppula, Nika Mansouri Ghiasi, Taha Shahroodi, Juan Gomez-Luna, and Onur Mutlu,
“SMASH: Co-designing Software Compression and Hardware-Accelerated Indexing for Efficient Sparse Matrix Operations”
Proceedings of the 52nd International Symposium on Microarchitecture (MICRO), Columbus, OH, USA, October 2019.
[Slides (pptx) (pdf)]
[Lightning Talk Slides (pptx) (pdf)]
[Poster (pptx) (pdf)]
[Lightning Talk Video (90 seconds)]
[Full Talk Lecture (30 minutes)]
Juan Gomez-Luna, Izzat El Hajj, Ivan Fernandez, Christina Giannoula, Geraldo F. Oliveira, and Onur Mutlu,
“Benchmarking a New Paradigm: Experimental Analysis and Characterization of a Real Processing-in-Memory System”
IEEE Access, 10 May 2022.
[arXiv version]
[PrIM Benchmarks Source Code]
[Slides (pptx) (pdf)]
[Long Talk Slides (pptx) (pdf)]
[Short Talk Slides (pptx) (pdf)]
[SAFARI Live Seminar Slides (pptx) (pdf)]
[SAFARI Live Seminar Video (2 hrs 57 mins)]
[Lightning Talk Video (3 minutes)]
[Short Talk Video (21 minutes)]
[1-hour Talk Video (58 minutes)]
Ivan Fernandez, Ricardo Quislant, Christina Giannoula, Mohammed Alser, Juan Gómez-Luna, Eladio Gutiérrez, Oscar Plata, and Onur Mutlu,
“NATSA: A Near-Data Processing Accelerator for Time Series Analysis”
Proceedings of the 38th IEEE International Conference on Computer Design (ICCD), Virtual, October 2020.
[Slides (pptx) (pdf)]
[Talk Video (10 minutes)]
[Source Code]
Core PhD Contributions with NTUA:
High-Performance and Balanced Parallel Graph Coloring on Multicore Platforms Journal of Supercomputing 2022 Christina Giannoula, Athanasios Peppas, Georgios Goumas, Nectarios Koziris
An Adaptive Concurrent Priority Queue for NUMA Architectures CF 2019 Foteini Strati*, Christina Giannoula*, Dimitrios Siakavaras, Georgios Goumas, Nectarios Koziris * Joint first authors
Combining HTM with RCU to Speed up Graph Coloring on Multicore Platforms ISC 2018 Christina Giannoula, Georgios Goumas, Nectarios Koziris
