Naos: Serialization-free RDMA networking in Java
Konstantin Taranov, Rodrigo Bruno, Gustavo Alonso, Torsten Hoefler
Abstract
Managed languages such as Java and Scala do not allow developers to directly access heap objects. As a result, to send on-heap data over the network, it has to be explicitly converted to byte streams before sending and converted back to objects after receiving. The technique, also known as object serialization/deserialization, is an expensive procedure limiting the performance of JVM-based distributed systems as it induces additional memory copies and requires data transformation resulting in high CPU and memory bandwidth consumption. This paper presents Naos, a JVM-based technique bypassing heap serialization boundaries that allows objects to be directly sent from a local heap to a remote one with minimal CPU involvement and over RDMA networks. As Naos eliminates the need to copy and transform objects, and enables asynchronous communication, it offers significant speedups compared to state-of-the-art serialization libraries. Naos exposes a simple high level API hiding the complexity of the RDMA protocol that transparently allows JVM-based systems to take advantage of offloaded RDMA networking.
- Modern RNICs support implicit on-demand paging (ODP) [17] that removes the need to register buffers. In our preliminary experiments, however, ODP performed worse than conventional explicit memory registration.
Ask about this paper
Your agent reads all of it.
Lune indexed this paper to the last equation, along with the top-tier papers that cite it. Ask a question and the answer quotes them.
Your agent calls
Luneget_paper_fulltext
Free to start. No credit card required.
Terminal
Install the CLIlune papers fulltext 950103ee-19b9-450e-bfe4-8130604a9628Cited by top-tier papers4
- Serialization/Deserialization-free State Transfer in Serverless WorkflowsFangming Lu, Xingda Wei, Zhuobin Huang, Rong Chen et al.EuroSys 2024 · 31 citations
- Cornflakes: Zero-Copy Serialization for Microsecond-Scale NetworkingDeepti Raghavan, Shreya Ravi, Gina Yuan, Pratiksha Thaker et al.SOSP 2023 · 18 citations
- TeraHeap: Reducing Memory Pressure in Managed Big Data FrameworksIacovos G. Kolokasis, Giannos Evdorou, Shoaib Akram, Christos Kozanitis et al.ASPLOS 2023 · 15 citations
- Zero-Change Object Transmission for Distributed Big Data AnalyticsMingyu Wu, Shuaiwei Wang, Haibo Chen, Binyu ZangUSENIX ATC 2022 · 7 citations
Builds on6
- sRDMA - Efficient NIC-based Authentication and Encryption for Remote Direct Memory AccessKonstantin Taranov, Benjamin Rothenberger, Adrian Perrig, Torsten HoeflerUSENIX ATC 2020 · 59 citations
- ReDMArk: Bypassing RDMA Security MechanismsBenjamin Rothenberger, Konstantin Taranov, Adrian Perrig, Torsten HoeflerUSENIX Security 2021 · 56 citations
- Optimus Prime: Accelerating Data Transformation in ServersArash Pourhabibi Zarandi, Siddharth Gupta, Hussein Kassir, Mark Sutherland et al.ASPLOS 2020 · 43 citations
- A Specialized Architecture for Object Serialization with Applications to Big Data AnalyticsJaeyoung Jang, Sungjun Jung, Sunmin Jeong, Jun Heo et al.ISCA 2020 · 32 citations
- CoRM: Compactable Remote Memory over RDMAKonstantin Taranov, Salvatore Di Girolamo, Torsten HoeflerSIGMOD 2021 · 18 citations
Related papers
- HeapBuffers: Why Not Just Using a Binary Serialization Format for Your Managed Memory?Daniele Bonetta, Júnior Löff, Matteo Basso, Walter BinderOOPSLA 2025 · 2 citations
- ODRP: On-Demand Remote Paging with Programmable RDMAZixuan Wang, Xingda Wei, Jinyu Gu, Hongrui Xie et al.NSDI 2025 · 7 citations
- JPDHeap: A JVM Heap Design for PM-DRAM MemoriesLitong You, Tianxiao Gu, Shengan Zheng, Jianmei Guo et al.DAC 2021 · 2 citations
- Semeru: A Memory-Disaggregated Managed RuntimeChenxi Wang, Haoran Ma, Shi Liu, Yuanqi Li et al.OSDI 2020 · 6 citations
- Maximizing the Benefit of RDMA at End HostsXiaoliang Wang, Hexiang Song, Cam-Tu Nguyen, Dongxu Cheng et al.INFOCOM 2021 · 8 citations
