Concurrent load test benchmark: Web Forms vs Blazor under pressure

[csharpfritz]

Summary

The sequential benchmarks from #424 (PR #464) show Blazor is 1.0x-2.7x faster for single-request response times. However, the real performance story -- especially for database-bound pages like ContosoUniversity Students -- requires concurrent load testing to capture throughput, latency under pressure, and resource utilization differences.

Motivation

Current benchmark results (single sequential requests):

App	Page	Framework	Blazor	Speedup
WingtipToys	Home	6.4ms	2.4ms	2.7x
WingtipToys	ProductList	8.3ms	3.6ms	2.3x
ContosoUniversity	Home	2.2ms	1.6ms	1.4x
ContosoUniversity	Students	6.5ms	6.3ms	~1.0x

The ContosoUniversity Students page shows almost no improvement because:

1. DB is the bottleneck -- both EF6 and EF Core hit the same LocalDB; framework overhead is negligible vs query time
2. Sequential requests don't stress the pipeline -- Web Forms' synchronous page lifecycle and ViewState serialization overhead only manifest under concurrent load
3. Blazor's async pipeline advantage is invisible -- async/await throughout the Blazor stack should handle concurrent requests more efficiently, but you can't see that with one request at a time

Proposed Work

1. Load test tooling

Use bombardier or similar HTTP load testing tool to run concurrent benchmarks:

• 10, 25, 50, 100 concurrent connections
• 30-second test duration per configuration
• Measure: requests/sec, avg latency, p50/p95/p99 latency, error rate

2. Test matrix

Same 4 apps and page sets as #424:

• WingtipToys (Framework) -- Home, ProductList, About
• AfterWingtipToys (Blazor) -- Home, ProductList, About
• ContosoUniversity (Framework) -- Home.aspx, Students.aspx, About.aspx
• AfterContosoUniversity (Blazor) -- Home, Students, About

3. Metrics to capture

• Throughput (req/sec) at each concurrency level
• Latency distribution (p50, p95, p99) under load
• Error rate under pressure
• Memory usage (process working set) during load test
• Thread pool saturation indicators

4. Expected findings

• Blazor should show significantly better throughput scaling due to async pipeline
• Web Forms thread-per-request model should saturate earlier
• Database-bound pages (Students) should show divergence at higher concurrency as framework overhead compounds
• ViewState serialization overhead in WingtipToys ProductList should amplify under load

5. Deliverables

• scripts/Run-LoadTests.ps1 -- orchestrator script
• Updated dev-docs/benchmarks/ with load test results, charts, and analysis
• Performance benchmark skill update with load test patterns

Acceptance Criteria

• All 4 apps benchmarked at 10/25/50/100 concurrent connections
• Results include throughput, latency percentiles, and error rates
• Report with comparison charts (throughput scaling curves, latency distributions)
• Analysis explaining where and why Blazor outperforms under load
• Skill updated with load test execution patterns

References

• Performance Test / Benchmark #424 -- Original performance benchmark issue
• PR perf: Performance benchmark suite - .NET Framework vs .NET 10 Blazor #464 -- Sequential benchmark implementation
• scripts/Run-Benchmarks.ps1 -- Existing benchmark infrastructure to extend
• .squad/skills/performance-benchmarks/SKILL.md -- Benchmark execution skill