阿部洋丈

• Sep. 2026

  Paper on a cooperative Kubernetes-TSN scheduler for latency-guaranteed applications presented at IEEE eScience 2026

• Jul. 2026

  Paper on SmartNIC-based memory monitoring for privilege escalation detection presented at PEARC '26

• Jul. 2026

  Paper on cooperative Kubernetes scheduling presented at IEEE ICMEW 2026

• Jun. 2026

  Two papers on network monitoring and admission control presented at ICCCI 2026

• May 2026

  Promoted to Professor, Institute of Systems and Information Engineering, University of Tsukuba

Date of birth

Nov. 21, 1976

Birthplace

Nagaoka, Niigata, Japan

My research designs system software that makes networked and distributed computing predictable and trustworthy. Current work spans resource orchestration for networked systems, measurement-based security monitoring, and machine-learning-assisted network control.

1. 01

   Systems Software

   • OS
   • Middleware
   • SmartNIC/kernel-level measurement

2. 02

   Computer Networks

   • Throughput estimation and optimization
   • Learning-based congestion control
   • Satellite communication

3. 03

   Distributed Systems

   • Resource orchestration (Kubernetes)
   • Latency-aware scheduling
   • Deterministic networking

4. 04

   Computer Security

   • Trusted execution environments
   • Remote attestation
   • Sandboxing & runtime monitoring

A selection of recent papers is featured below. The complete list of publications is available on the full list page.

Full publication list →

1. 01

   Takashi Yagawa, Tadanori Teruya, Kazuma Ohara, Kuniyasu Suzaki, Hirotake Abe

   Clarifying CPU Vendor’s Responsibilities for Remote Attestation with Intel SGX

   IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences

   vol.E109-A, no.3, pp.305–316, 2026

   Intel SGX is moving from EPID-based to ECDSA-based remote attestation, which lets third parties build their own verification environments with the Data Center Attestation Primitives (DCAP). This flexibility, however, blurs the responsibility boundary between Intel and those third parties. By systematically comparing the components of both attestation schemes, this paper shows that Intel's duty reduces to distributing signed data rather than vouching for the whole verification process, and identifies the Quoting Enclave (QE3) as the keystone of that boundary. Experiments on real hardware further confirm that a malicious QE3 is constrained by the hardware specification itself.

2. 02

   Yohei Kojima, Ryu Kazama, Hirotake Abe, Chunghan Lee

   RNN-based Congestion Control in the Linux Kernel

   The 9th International Workshop on GPU Computing and AI (GCA), held in conjunction with the 12th International Symposium on Computing and Networking (CANDARW)

   pp.130–136, 2024

   This paper implements RNN-CUBIC, a congestion control scheme that extends CUBIC with a recurrent neural network balancing throughput, delay, and packet loss, directly inside the Linux kernel. Running the predictor in kernel space removes context-switch overhead, and lightweight design choices such as integer-only arithmetic and an exponential-free softsign activation keep prediction cheap. Measured over the real Internet between the University of Tsukuba and the AWS Tokyo region, RNN-CUBIC improves average throughput by about 20% over CUBIC while adding less than 1 ms per prediction.

3. 03

   Teruto Endo, Hirotake Abe, Mizuki Oka

   Toward Automatic Generation of Diverse Congestion Control Algorithms through Co-evolution with Simulation Environments

   ALIFE 2022: The 2022 Conference on Artificial Life (MIT Press)

   Paper No: isal_a_00515, 33; 8 pages, 2022

   Congestion control algorithms underpin today's Internet, yet their evolution is slow to keep pace with rapidly changing networks and applications. Inspired by POET (Pair-wised Open-Ended Trailblazer), this paper generates new congestion control algorithms automatically by letting algorithms and network environments co-evolve. It introduces a Quality Diversity-based evolution procedure and examines the diverse algorithms it produces.

4. 04

   R. N. S. Widodo, H. Ohtsuji, E. Hayashi, E. Yoshida, H. Abe, K. Kato

   NVKVS: Non-Volatile Memory Optimized Key-Value Separated LSM-Tree

   Journal of Information Processing

   vol.30, pp.332–342, 2022

   Key-value stores are essential infrastructure for cloud services and distributed file systems, and speeding them up is an active pursuit worldwide. Building on the key-value separated LSM-tree proposed by the University of Wisconsin, this paper inserts non-volatile memory between main memory and SSD, tuned to commercially available Intel Optane DCPM. The design suppresses write amplification, the main cause of performance loss, and substantially improves scalability over the original study.

5. 05

   R. N. S. Widodo, H. Abe, K. Kato

   Hadoop Data Reduction Framework: Applying Data Reduction at the DFS Layer

   IEEE Access

   vol.9, pp.152704–152717, 2021

   Hadoop's distributed file system could benefit greatly from data reduction, but conventional approaches embed it either in the application, leaving generic tools unsupported, or in the local file system, wasting bandwidth on unreduced traffic over the network. This paper proposes a third option: applying data reduction at the DFS layer itself. Extending the design first presented at ACM APSys 2020, it adds block mirroring to further cut background traffic, and evaluates the system thoroughly on testbeds interconnected at 2.5 Gbps and 10 Gbps, demonstrating its effectiveness.