Empirical Software Research

LAYERLEE Engineering Lab

Software engineering is an empirical discipline. The Engineering Lab hosts reproducible benchmarks, hardware latency experiments, and architecture investigations conducted by Bipin Baral.

Zero Fabricated Results·Explicit Hardware & OS Disclosures·Reproducible Commands

EXP-001

Completed Benchmark

database

Read Associated Deep-Dive

SQLite WAL Mode vs In-Memory Execution Latency Under High-Concurrency Reads

HypothesisEnabling Write-Ahead Logging (WAL) in SQLite provides near in-memory read throughput while maintaining atomic persistence, scaling to 10,000 queries/second without lock contention.

Operating SystemmacOS 15 (Darwin 24.x) / Linux 6.6 LTS

Runtime / FrameworkNode.js v20.18 / better-sqlite3 11.x

Hardware SetupApple M-Series / 16GB Unified RAM / NVMe SSD

Methodology:

Benchmarked 50,000 indexed point lookups across concurrent worker threads comparing default rollback journal mode (DELETE) vs WAL mode (PRAGMA journal_mode=WAL) with PRAGMA synchronous=NORMAL.

Dataset / Workload:

1,000,000 synthetic indexed records (UUID, timestamp, JSON payload).

Reproducibility Commands

PRAGMA journal_mode = WAL;
PRAGMA synchronous = NORMAL;
PRAGMA temp_store = MEMORY;
PRAGMA cache_size = -64000;

Observed ResultsWAL mode delivered a 4.8x improvement in concurrent read throughput over rollback journal mode, maintaining < 0.2ms P99 query latency during heavy concurrent write operations.

Limitations: Tested on local NVMe storage. Distributed replication over network storage (NFS/EFS) introduces filesystem sync bottlenecks not evaluated in this local benchmark.

EXP-002

Completed Benchmark

backend

Read Associated Deep-Dive

NestJS Fastify Adapter vs Express HTTP Throughput in JSON Serialization

HypothesisSwitching NestJS HTTP runtime from the default platform-express to platform-fastify with fast-json-stringify yields a 2.5x increase in raw request throughput on JSON API payloads.

Operating SystemUbuntu 22.04 LTS (Docker Container)

Runtime / FrameworkNode.js 20.18 LTS / NestJS 10.4.x

Hardware Setup4 vCPU / 8GB RAM Cloud VPS

Methodology:

Executed Autocannon load tests (100 concurrent connections, 30-second duration) targeting identical health check and nested JSON serialization endpoints.

Dataset / Workload:

Nested JSON payload (50 keys, array of 20 items, ~4KB per response).

Reproducibility Commands

autocannon -c 100 -d 30 -p 10 http://localhost:3000/api/benchmark/express
autocannon -c 100 -d 30 -p 10 http://localhost:3000/api/benchmark/fastify

Observed ResultsFastify adapter handled 38,400 req/sec compared to Express's 15,200 req/sec with a 62% reduction in memory pressure during peak load.

Limitations: Raw HTTP benchmarks do not account for database I/O bottlenecks in typical CRUD operations where database latency dominates response time.

EXP-003

Completed Benchmark

operating systems

Read Associated Deep-Dive

PlayStation Orbis OS Syscall Filter Analysis (FreeBSD 9.0 Base)

HypothesisSony's Orbis OS kernel restricts unprivileged userland access to approximately 65 standard FreeBSD system calls, isolating game sandboxes via mandatory access control (MAC) policies.

Operating SystemFreeBSD 9.0 Kernel Reference / Orbis OS Security Model

Runtime / FrameworkLow-level system architecture analysis

Hardware Setupx86-64 AMD Jaguar APU architecture

Methodology:

Cross-referenced public security disclosures, FreeBSD kernel source diffs, and game engine sandbox limitations to map the Orbis OS kernel security architecture.

Dataset / Workload:

Standard FreeBSD 9.0 syscall table (530+ syscalls) compared against Orbis SDK documentation.

Reproducibility Commands

grep -rn 'sysent' sys/kern/makesyscalls.sh
objdump -d -j .text game_runtime.elf | grep -E 'syscall|int 0x80'

Observed ResultsMapped the multi-layer security architecture isolating the WebKit UI layer from the privileged game kernel via proprietary SAMU cryptographic co-processors.

Limitations: Based on reverse-engineering disclosures and publicly released security papers; direct kernel source is proprietary to Sony Interactive Entertainment.

EXP-004

Research Proposal

database

Read Associated Deep-Dive

PostgreSQL B-Tree vs GIN Index Query Plan Execution on 10M Row JSONB

HypothesisUsing a specialized GIN expression index with jsonb_path_ops reduces index size by 40% while matching single-key query speed of a dedicated B-Tree functional index.

Operating SystemDebian 12 / Docker

Runtime / FrameworkPostgreSQL 16.4

Hardware Setup8 vCPU / 16GB RAM / NVMe

Methodology:

Proposed automated SQL benchmark generating 10 million telemetry JSONB rows and comparing index build duration, storage footprint, and EXPLAIN ANALYZE execution time for containment queries.

Dataset / Workload:

10,000,000 synthetic IoT telemetry records.

Reproducibility Commands

CREATE INDEX idx_telemetry_gin ON telemetry USING gin (data jsonb_path_ops);
EXPLAIN (ANALYZE, BUFFERS, TIMING) SELECT * FROM telemetry WHERE data @> '{"device_id": 42}';

Limitations: Experiment proposal pending full cluster setup. Benchmarks will be published once complete.