Parul Sohal
PhD · Computer Systems
I build systems that measure, analyze, and improve performance. My work focuses on understanding how applications interact with shared resources and using that insight to make computing systems more efficient, predictable, and scalable. Through a data-driven approach, I translate performance measurements into actionable insights and quantifiable results.
Projects
Profile-driven Performance Management Framework (ProPer)
Profile-driven Performance Management · Cloud
Architected a profile-driven resource management framework that predicts application phase changes at function boundaries and proactively repartitions cache and memory bandwidth, cutting worst-case slowdown under heavy contention from 6× to 2.5× at under 0.3% profiling overhead.
View project →Envelope-aWare Predictive Model (E-WarP)
Embedded Systems · ARM · Real-Time
Deployed an end-to-end, hybrid software/hardware resource management framework for heterogeneous embedded systems with a microsecond-resolution profiler at <1% overhead, integrating MemGuard software bandwidth regulation for CPU cores with ARM QoS hardware enforcement for accelerators within the Jailhouse partitioning hypervisor.
View project →Unikernel Linux (UKL)
OS · Kernel · Systems
Collaborated to develop the initial Linux kernel target for a unikernel-style execution model, enabling a single application to link directly with the kernel and run with supervisor privileges in a unified address space.
View project →Publications
Unikernel Linux (UKL)
Profile-Driven Memory Bandwidth Management for Accelerators and CPUs in QoS-Enabled Platforms
A Closer Look at Intel Resource Director Technology (RDT)
E-WarP: A System-Wide Framework for Memory Bandwidth Profiling and Management
Unikernels: The Next Stage of Linux's Dominance
Questions That Led Somewhere Interesting
Small engineering problems often teach more than the final result. This section is a collection of short notes on interesting challenges I encountered while building systems — from debugging obscure performance issues to writing tools, benchmarks, and experiments that didn't make it into the papers but shaped the work.
Browse the notes →Technical Skills
Systems
- Hypervisors
- Resource Isolation
- Linux Systems Programming
- Kernel Development
- Real-time Systems
- Workload Phase Detection
- Memory Contention Analysis
Hardware & Platforms
- Intel RDT (CAT, MBA)
- ARM QoS
- Cache Partitioning
- DRAM Bandwidth Regulation
- PMUs
- Multi-core x86 (Cascade Lake, Ice Lake)
- ARM-based SoCs (NXP S32V234)
Languages & Tools
- C, Python, Bash
- Perf, SystemTap
- Benchmarking, Git
ML / AI
- PyTorch
- Codex
- Claude Code
Let's work together
I’m currently exploring opportunities in performance engineering, cloud infrastructure, distributed systems, and ML infrastructure. My work focuses on understanding and optimizing the interaction between software and hardware to deliver measurable performance improvements. If you’re working on systems where efficiency, scalability, and performance matter, I’d love to connect.
Contact me