Kanishk Agarwal
Backend Engineer specializing in low-latency systems, concurrency, and distributed execution engines.
I build production-grade infrastructure, handling bottlenecks, thread synchronization, and system constraints.
85%P99 Latency Reduction
100k+Events/sec Throughput
ZeroData Races / Deadlocks
Featured System
C++ Concurrent Task Scheduler
PROBLEM & CONSTRAINTS
Standard thread spawning overhead was creating a 40ms bottleneck on high-frequency task dispatches. The system required a bounded-memory pool capable of absorbing temporary backpressure without causing thread starvation or excessive context switching.
SYSTEM DESIGN
- Fixed-size worker pool sized to `std::thread::hardware_concurrency()`
- Lock-free ring buffer concepts applied to submission queues to minimize contention
- Condition variable signaling for bounded idle worker wakeups
- Type-erased packaged tasks returning `std::future` for asynchronous resolution
TRADE-OFFS & MEASURED RESULTS
Opted for a single centralized queue over work-stealing queues due to implementation complexity vs L1 cache coherency overhead at our specific scale. Result: Eliminated arbitrary thread creation overhead, dropping P99 dispatch latency from 40ms to 1.2ms under heavy burst load.
Technical Deep Dives
SDE-2 Competencies
- Multithreaded ExecutionDesigning thread pools, safe memory models, and minimizing context switches in bounded resources.
- State Machine OrchestrationDirecting state, idempotency, and retry layers for LLM and workflow automation logic.