Vahab Jabrayilov
Ph.D. Candidate, Department of Computer Science, Columbia University
Systems engineer focused on ML performance: accelerators (TPU, GPU), low-latency networking, and scheduling.
I’m a Ph.D. candidate in Computer Science at Columbia University, advised by Prof. Kostis Kaffes. I work on ML systems, with a focus on making the infrastructure under AI workloads faster and more reliable. My work so far has covered:
- Accelerators: optimizing TPU fleet utilization through runtime and compiler performance work at Google, and adding fault isolation to GPU sharing with Fence
- Networking: enabling low-latency networking in public cloud VMs with Machnet at Microsoft Research
- Scheduling: enabling microsecond-scale VM scheduling with Rorke
Work Experience
Student Researcher @ Google Now
May 2026 - Present
TPU Runtime team · New York, NY and Sunnyvale, CA, USA
Optimize TPU fleet utilization with runtime and compiler performance work.
Mentors:
Ofir Weisse and Buddh Prakash
Graduate Research Assistant @ Columbia University Now
Aug 2023 - Present
New York, NY, USA
- Built Fence, a runtime that adds fault isolation to dynamic spatial sharing on unmodified NVIDIA GPUs for LLM serving.
- Built Machnet, a C++ DPDK messaging library for public cloud VMs that handles 750K RPS on Azure at 61us tail latency.
- Built Rorke, a microsecond-scale VM scheduler in Rust and eBPF.
Advisor:
Prof. Kostis Kaffes
Research Software Engineer Intern @ Microsoft Research
Jun 2024 - Sep 2024
Networking Research Group · Redmond, WA, USA
Built a Rust DPDK packet processor that handles millions of flow logs/sec with deterministic sub-ms latency for Azure's real-time data ingestion pipeline.
Mentors:
Sathiya Kumaran Mani and Srikanth Kandula
Research Intern @ EPFL
Jun 2023 - Aug 2023
Lausanne, Switzerland
Analyzed metastable failures in distributed systems and designed mitigations.
Advisor:
Prof. Sanidhya Kashyap
Research Intern @ Penn State University
State College, PA, USA
Built Replicant, a MultiPaxos-based key-value store for strongly consistent cloud replication.
Advisor:
Abutalib Aghayev
Selected Publications
ASPLOS 2026
Enabling Fast Networking in the Public Cloud
Presents Machnet, a C++ DPDK messaging library for public cloud VMs.
PDFACMP99 CONF TalkNYC Systems Talk
Abstract
Despite a decade of research, most high-performance userspace network stacks remain impractical for public cloud tenants developing their applications atop Virtual Machines (VMs). We identify two root causes: (1) reliance on specialized NIC features (e.g., flow steering, deep buffers) absent in commodity cloud vNICs, and (2) rigid execution models ill-suited to diverse application needs. We present Machnet, a high-performance and flexible userspace network stack designed for public cloud VMs. Machnet uses only a minimal set of vNIC features that any major cloud provider supports. It also relies on a microkernel architecture to enable flexible application execution. We evaluate Machnet across three major public clouds and on production-grade applications, including a key-value store, an HTTP server, and a state-machine replication system. We release Machnet at https://github.com/microsoft/machnet.
BibTeX
@inproceedings{10.1145/3779212.3790158,
author = {Sanaee, Alireza and Jabrayilov, Vahab and Marinos, Ilias and Shahinfar, Farbod and Saxena, Divyanshu and Antichi, Gianni and Kaffes, Kostis},
title = {Enabling Fast Networking in the Public Cloud},
booktitle = {Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2},
pages = {679--696},
year = {2026},
doi = {10.1145/3779212.3790158}
}
Under Review
Fence: Safe Dynamic Spatial Sharing for GPUs
Adds fault isolation to dynamic spatial sharing on unmodified NVIDIA GPUs for LLM serving.
Under Review
Rorke: Black-box Virtual Machine Scheduling for the Public Cloud
A microsecond-scale VM scheduler in Rust and eBPF.
NeurIPS MLForSys Workshop 2025
An Expert in Residence: LLM Agents for Always-On Operating System Tuning
Uses LLM agents to tune operating systems continuously.
PDFNeurIPS
Abstract
Classical machine-learning auto-tuners for OS control struggle with semantic gaps, brittle rewards, and unsafe exploration. We introduce an online, LLM-driven agent that emulates expert reasoning for continuous OS optimization. When tuning the Linux Completely Fair Scheduler’s hyperparameters, the agent outperforms Bayesian optimization by 5% in single-parameter tuning, 7.1% in two-parameter co-tuning, and a human expert by 2.98% overall, while converging faster and adapting more quickly to workload changes. When application counters are unavailable, system-level proxies (e.g., Instructions Per Cycle (IPC)) preserved tail latency in our setup. Putting this together, we propose adopting the Model Context Protocol (MCP) for tool/resource discovery and invocation and a logging channel; on top of that, we propose adding transactional apply–commit–revert, host-mediated approval gates, and policy controls in the OS-tuning server and host to ensure safe, auditable operation. Our results and reference design suggest a practical path toward safe, self-adapting OS control.
BibTeX
@inproceedings{liargkovas2025expert,
author = {Liargkovas, Georgios and Jabrayilov, Vahab and Franke, Hubertus and Kaffes, Kostis},
title = {An Expert in Residence: LLM Agents for Always-On Operating System Tuning},
booktitle = {NeurIPS 2025 Workshop on Machine Learning for Systems},
year = {2025}
}
VLDB 2025
HoliPaxos: Towards More Predictable Performance in State Machine Replication
Makes state machine replication performance more predictable.
PDFVLDB
Abstract
State machine replication (SMR) algorithms ensure redundancy in critical systems and, as a result, underpin fault-tolerant distributed databases. Good SMR protocol performance is essential for capacity planning and meeting desired performance objectives. However, many implementations of popular SMR algorithms, such as MultiPaxos and Raft, have issues that make their performance unpredictable. This unpredictability often arises from certain “bolt-on” additions to core protocols, such as external failure detectors and replication log compaction. In this paper, we argue that tighter integration of such traditionally ad-hoc mechanisms with the core replication protocols can stabilize performance, making the solutions more reliable and more accessible to accurate capacity planning. Moreover, we show that these integrations can be nondisruptive for the underlying consensus algorithm, resulting in systems that preserve the simplicity and safety of traditional single-leader consensus-based SMR. To that order, we integrate the failure and slowdown detectors inside the SMR and achieve better performance and faster fail-over under various network partitions and node slowdown events. We also illustrate that tight integration of replication log management, pruning, and snapshotting can reduce memory and CPU usage while avoiding performance fluctuations associated with traditional log compaction and cleanup approaches.
BibTeX
@article{DBLP:journals/pvldb/LiangJAC25,
author = {Liang, Zhiying and Jabrayilov, Vahab and Aghayev, Abutalib and Charapko, Aleksey},
title = {HoliPaxos: Towards More Predictable Performance in State Machine Replication},
journal = {Proc. {VLDB} Endow.},
volume = {18},
number = {8},
pages = {2505--2518},
year = {2025}
}
All publications (10)
ASPLOS 2026
Enabling Fast Networking in the Public Cloud
Presents Machnet, a C++ DPDK messaging library for public cloud VMs.
PDFACMP99 CONF TalkNYC Systems Talk
Abstract
Despite a decade of research, most high-performance userspace network stacks remain impractical for public cloud tenants developing their applications atop Virtual Machines (VMs). We identify two root causes: (1) reliance on specialized NIC features (e.g., flow steering, deep buffers) absent in commodity cloud vNICs, and (2) rigid execution models ill-suited to diverse application needs. We present Machnet, a high-performance and flexible userspace network stack designed for public cloud VMs. Machnet uses only a minimal set of vNIC features that any major cloud provider supports. It also relies on a microkernel architecture to enable flexible application execution. We evaluate Machnet across three major public clouds and on production-grade applications, including a key-value store, an HTTP server, and a state-machine replication system. We release Machnet at https://github.com/microsoft/machnet.
BibTeX
@inproceedings{10.1145/3779212.3790158,
author = {Sanaee, Alireza and Jabrayilov, Vahab and Marinos, Ilias and Shahinfar, Farbod and Saxena, Divyanshu and Antichi, Gianni and Kaffes, Kostis},
title = {Enabling Fast Networking in the Public Cloud},
booktitle = {Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2},
pages = {679--696},
year = {2026},
doi = {10.1145/3779212.3790158}
}
Under Review
Fence: Safe Dynamic Spatial Sharing for GPUs
Adds fault isolation to dynamic spatial sharing on unmodified NVIDIA GPUs for LLM serving.
Under Review
Rorke: Black-box Virtual Machine Scheduling for the Public Cloud
A microsecond-scale VM scheduler in Rust and eBPF.
NeurIPS MLForSys Workshop 2025
An Expert in Residence: LLM Agents for Always-On Operating System Tuning
Uses LLM agents to tune operating systems continuously.
PDFNeurIPS
Abstract
Classical machine-learning auto-tuners for OS control struggle with semantic gaps, brittle rewards, and unsafe exploration. We introduce an online, LLM-driven agent that emulates expert reasoning for continuous OS optimization. When tuning the Linux Completely Fair Scheduler’s hyperparameters, the agent outperforms Bayesian optimization by 5% in single-parameter tuning, 7.1% in two-parameter co-tuning, and a human expert by 2.98% overall, while converging faster and adapting more quickly to workload changes. When application counters are unavailable, system-level proxies (e.g., Instructions Per Cycle (IPC)) preserved tail latency in our setup. Putting this together, we propose adopting the Model Context Protocol (MCP) for tool/resource discovery and invocation and a logging channel; on top of that, we propose adding transactional apply–commit–revert, host-mediated approval gates, and policy controls in the OS-tuning server and host to ensure safe, auditable operation. Our results and reference design suggest a practical path toward safe, self-adapting OS control.
BibTeX
@inproceedings{liargkovas2025expert,
author = {Liargkovas, Georgios and Jabrayilov, Vahab and Franke, Hubertus and Kaffes, Kostis},
title = {An Expert in Residence: LLM Agents for Always-On Operating System Tuning},
booktitle = {NeurIPS 2025 Workshop on Machine Learning for Systems},
year = {2025}
}
VLDB 2025
HoliPaxos: Towards More Predictable Performance in State Machine Replication
Makes state machine replication performance more predictable.
PDFVLDB
Abstract
State machine replication (SMR) algorithms ensure redundancy in critical systems and, as a result, underpin fault-tolerant distributed databases. Good SMR protocol performance is essential for capacity planning and meeting desired performance objectives. However, many implementations of popular SMR algorithms, such as MultiPaxos and Raft, have issues that make their performance unpredictable. This unpredictability often arises from certain “bolt-on” additions to core protocols, such as external failure detectors and replication log compaction. In this paper, we argue that tighter integration of such traditionally ad-hoc mechanisms with the core replication protocols can stabilize performance, making the solutions more reliable and more accessible to accurate capacity planning. Moreover, we show that these integrations can be nondisruptive for the underlying consensus algorithm, resulting in systems that preserve the simplicity and safety of traditional single-leader consensus-based SMR. To that order, we integrate the failure and slowdown detectors inside the SMR and achieve better performance and faster fail-over under various network partitions and node slowdown events. We also illustrate that tight integration of replication log management, pruning, and snapshotting can reduce memory and CPU usage while avoiding performance fluctuations associated with traditional log compaction and cleanup approaches.
BibTeX
@article{DBLP:journals/pvldb/LiangJAC25,
author = {Liang, Zhiying and Jabrayilov, Vahab and Aghayev, Abutalib and Charapko, Aleksey},
title = {HoliPaxos: Towards More Predictable Performance in State Machine Replication},
journal = {Proc. {VLDB} Endow.},
volume = {18},
number = {8},
pages = {2505--2518},
year = {2025}
}
ICDCS 2025
Practical Considerations for Implementing State Machine Replication in the Cloud
PDFNSF PAR
Abstract
State Machine Replication (SMR) protocols form the backbone of many distributed systems. Enterprises and startups increasingly build their distributed systems on the cloud due to its many advantages, such as scalability and cost-effectiveness. Due to their prevalence in systems, the practical aspects of SMR algorithms, such as efficiency and performance, become hugely important. These practical considerations can impact capacity planning, deployment strategies, and, ultimately, the cost of running stateful systems in the cloud. In this paper, we consider various practical choices that impact the performance and efficiency of state machine replication in the cloud. To that order, we design a language-agnostic Multi-Paxos-based state machine architecture and implement it in several languages popular for cloud deployment. In the process, we investigate the impact of threading, communication, and memory management models on replicated state machines’ performance and resource efficiency. We also examine the high-level implications of virtualization on the performance of SMR implementations in various languages. We present our findings as a collection of practical lessons backed by our experimental data and analysis.
BibTeX
@inproceedings{liang2025practical,
author = {Liang, Zhiying and Jabrayilov, Vahab and Aghayev, Abutalib and Charapko, Aleksey},
title = {Practical Considerations for Implementing State Machine Replication in the Cloud},
booktitle = {IEEE 45th International Conference on Distributed Computing Systems (ICDCS)},
pages = {791--801},
year = {2025}
}
ACM SoCC 2024
Rethinking the Networking Stack for Serverless Environments: A Sidecar Approach
PDFACM
Abstract
Serverless platforms rely on legacy networking stacks for communication and data movement. We quantitatively analyze the performance of these stacks and show their mismatch with highly consolidated, virtualized modern serverless environments, focusing on Firecracker, the most common serverless virtualization framework. As serverless applications grow in complexity and interaction, the resulting network bottleneck is a prime source of user-perceived, end-to-end latency. In this paper, we present a detailed vision of a new, sidecar-based networking stack for serverless environments. Our primary design goal is to provide low-overhead networking while maintaining existing security guarantees. We outline the research challenges in both the control and the data plane that the community needs to tackle before such a sidecar architecture can be used in practice.
BibTeX
@article{10.1145/3698038.3698561,
author = {Seshagiri, Vishwanath and Gupta, Abhinav and Jabrayilov, Vahab and Wildani, Avani and Kaffes, Kostis},
title = {Rethinking the Networking Stack for Serverless Environments: A Sidecar Approach},
year = {2024},
isbn = {9798400712869},
publisher = {Association for Computing Machinery},
booktitle = {Proceedings of the 2024 ACM Symposium on Cloud Computing},
pages = {213–222},
series = {SoCC '24}
}
arXiv 2024
MultiPaxos Made Complete
arXiv
Abstract
MultiPaxos, while a fundamental Replicated State Machine algorithm, suffers from a dearth of comprehensive guidelines for achieving a complete and correct implementation. This deficiency has hindered MultiPaxos' practical utility and adoption and has resulted in flawed claims about its capabilities. Our paper aims to bridge the gap between MultiPaxos' complexity and practical implementation through a meticulous and detailed design process spanning more than a year. It carefully dissects each phase of MultiPaxos and offers detailed step-by-step pseudocode -- in addition to a complete open-source implementation -- for all components, including the leader election, the failure detector, and the commit phase. The implementation of our complete design also provides better performance stability, resource usage, and network partition tolerance than naive MultiPaxos versions. Our specification includes a lightweight log compaction approach that avoids taking repeated snapshots, significantly improving resource usage and performance stability. Our failure detector, integrated into the commit phase of the algorithm, uses variable and adaptive heartbeat intervals to settle on a better leader under partial connectivity and network partitions, improving liveness under such conditions.
BibTeX
@article{liang2024multipaxoscomplete,
title = {MultiPaxos Made Complete},
author = {Liang, Zhiying and Jabrayilov, Vahab and Charapko, Aleksey and Aghayev, Abutalib},
year = {2024},
eprint = {2405.11183},
archivePrefix = {arXiv},
primaryClass = {cs.DC}
}
arXiv 2024
The Cost of Garbage Collection for State Machine Replication
arXiv
Abstract
State Machine Replication (SMR) protocols form the backbone of many distributed systems. Enterprises and startups increasingly build their distributed systems on the cloud due to its many advantages, such as scalability and cost-effectiveness. One of the first technical questions companies face when building a system on the cloud is which programming language to use. Among many factors that go into this decision is whether to use a language with garbage collection (GC), such as Java or Go, or a language with manual memory management, such as C++ or Rust. Today, companies predominantly prefer languages with GC, like Go, Kotlin, or even Python, due to ease of development; however, there is no free lunch: GC costs resources (memory and CPU) and performance (long tail latencies due to GC pauses). While there have been anecdotal reports of reduced cloud cost and improved tail latencies when switching from a language with GC to a language with manual memory management, so far, there has not been a systematic study of the GC overhead of running an SMR-based cloud system. This paper studies the overhead of running an SMR-based cloud system written in a language with GC. To this end, we design from scratch a canonical SMR system -- a MultiPaxos-based replicated in-memory key-value store -- and we implement it in C++, Java, Rust, and Go. We compare the performance and resource usage of these implementations when running on the cloud under different workloads and resource constraints and report our results. Our findings have implications for the design of cloud systems.
BibTeX
@article{liang2024costgarbagecollectionstate,
title = {The Cost of Garbage Collection for State Machine Replication},
author = {Liang, Zhiying and Jabrayilov, Vahab and Charapko, Aleksey and Aghayev, Abutalib},
year = {2024},
eprint = {2405.11182},
archivePrefix = {arXiv},
primaryClass = {cs.DC}
}
SRMC 2022
An Empirical Analysis of IDS Approaches in Container Security
IEEE
BibTeX
@article{9973138,
author = {Sever, Yigit and Ekinci, Goktug and Dogan, Adnan Harun and Alparslan, Bugra and Gurbuz, Abdurrahman Said and Jabrayilov, Vahab and Angin, Pelin},
booktitle = {2022 International Workshop on Secure and Reliable Microservices and Containers (SRMC)},
title = {An Empirical Analysis of IDS Approaches in Container Security},
year = {2022},
pages = {18-26}
}
Talks
P99 CONF 2026
How We Hit Bare-Metal Tail Latency Inside Cloud VMs: Machnet
Covers Machnet's design and tradeoffs.
ASPLOS 2026
Enabling Fast Networking in the Public Cloud
Conference presentation of the Machnet paper.
Education
Columbia University, United States
Sep. 2023 - Present (Expected 2027)Ph.D. in Computer Science
Advised by Prof. Kostis Kaffes
Columbia University, United States
Sep. 2023 - 2025M.Sc. in Computer Science
Middle East Technical University, Turkey
2019 - 2023B.Sc. in Computer Engineering
Honors
ACM ICPC semi-final Honorable Mention
2018
Rank 1 of ~70,000, Azerbaijan National University Entrance Exam
2017