Major Optimizations

Benchmark Results: OxyPlot.DotNetAndroid

All benchmarks run on the host CLR (.NET 8, Windows) using BenchmarkDotNet 0.14.0 with [MemoryDiagnoser].
Android JNI overhead cannot be measured off-device. Relative improvements are valid; absolute numbers will differ on ART.

FontMetricsBenchmark

Measures the cost of allocating a new FontMetrics-equivalent object per call (baseline, mirrors Paint.GetFontMetrics() no-arg overload) vs. filling a cached instance in-place (mirrors Paint.GetFontMetrics(Paint.FontMetrics) overload).

Method	Mean	Error	StdDev	Ratio	Gen0	Allocated	Alloc Ratio
`GetFontMetrics_Allocating` (baseline)	2.944 ns	0.100 ns	0.200 ns	1.00	0.0048	40 B	1.00
`GetFontMetrics_Cached` (Fix 1)	0.467 ns	0.048 ns	0.045 ns	0.16	–	0 B	0.00

Key finding: Caching the FontMetrics instance eliminates 40 bytes of allocation per call (100% reduction) and reduces call cost by 6.3×. At 10 views × 60 FPS with text on every frame, this removes ~24,000 allocations/second.

PathBuildBenchmark

Measures the cost of resetting and re-populating a path-like list structure (proxy for Android.Graphics.Path.Reset() + LineTo).

Method	PointCount	Mean	Ratio	Gen0	Allocated
`BuildPath_ReuseList` (Fix 2 pattern)	100	127.1 ns	1.00	–	0 B
`BuildPath_NewList` (naive)	100	130.2 ns	1.03	0.1023	856 B
`BuildPath_ReuseList`	500	583.5 ns	1.00	–	0 B
`BuildPath_NewList`	500	681.5 ns	1.17	0.4845	4,056 B
`BuildPath_ReuseList`	1000	1,163.6 ns	1.00	–	0 B
`BuildPath_NewList`	1000	1,310.3 ns	1.13	0.9680	8,104 B

Key finding: Reusing a single path instance eliminates all managed allocations and is 3–17% faster depending on point count. For 500-point series (the stress test scenario), reuse saves 4 KB per draw call.

RenderContextBenchmark

Drives a no-op MockRenderContext through a simulated OnDraw cycle: axis rectangles + 500-point line series + 18 axis labels.

Method	Mean	Error	StdDev	Gen0	Allocated
`SimulatedFrame`	446.6 ns	10.48 ns	30.40 ns	0.0715	600 B

Key finding: A complete simulated frame (no Android JNI) allocates only 600 bytes at the OxyPlot+render-context layer. This is well within the target of <4 KB/frame. The remaining allocation is from OxyPlot's internal axis/series rendering pipeline.

PlotModelUpdateBenchmark

Calls IPlotModel.Update() on a PlotModel with one LineSeries of 500 points.

Method	Mean	Error	StdDev	Gen0	Allocated
`Update_WithData`	2,482.5 ns	45.63 ns	42.68 ns	0.1793	1.48 KB
`Update_WithoutData`	573.1 ns	9.95 ns	12.58 ns	0.1640	1.34 KB

Key finding: The OxyPlot model update itself allocates ~1.3–1.5 KB per call regardless of updateData. This is OxyPlot library-internal overhead. The updateData: false path (used in the hot OnDraw loop after the first frame) is 4.3× faster.

Phase 3 Stress Test Results

Test: StressTest_10Views_60Hz_5Minutes — 10 models × 60 Hz × 300 seconds = 18,000,000 simulated renders via MockRenderContext.

Baseline (before fixes)

Metric	Value
Elapsed	~3.71 s (simulated time)
Avg frame time	0.206 ms
Gen-0 GCs	272
Gen-1 GCs	1
Gen-2 GCs	0
Heap growth	5,480 KB

After fixes (Fix 1–5 applied)

Metric	Value	vs Baseline
Elapsed	~3.64 s	−1.8%
Avg frame time	0.202 ms	−1.9%
Gen-0 GCs	273	+0.4% (within noise)
Gen-1 GCs	1	same
Gen-2 GCs	0	same
Heap growth	5,576 KB	+1.7% (within noise)
Test result	✅ PASSED	—

Note: The stress test uses MockRenderContext (no-op), so Android JNI costs (including Paint.GetFontMetrics() JNI allocation) are not exercised. The GC numbers in the stress test reflect OxyPlot's own internal allocations only. The FontMetrics fix benefit will be visible on device as eliminated JNI-backed object allocations, not in this host-CLR benchmark.

Before / After Summary Table

Benchmark / Metric	Before	After	Improvement
`GetFontMetrics` allocated bytes/call	40 B	0 B	100% reduction
`GetFontMetrics` mean call time	2.94 ns	0.47 ns	6.3× faster
`BuildPath` allocated bytes/op (500 pts)	4,056 B	0 B	100% reduction
`SimulatedFrame` allocated bytes/frame	600 B	600 B	(no-op RC — host baseline)
Stress test gen-0 GCs / 5 min	272	273	within noise
Stress test gen-2 GCs / 5 min	0	0	✅
Stress test heap growth	5,480 KB	5,576 KB	within noise
Build passes	✅	✅	—
New compiler warnings	0	0	—

Acceptance Criteria Status

Criterion	Target	Result
Allocated bytes per `DrawText` call (host proxy)	0 B after fix	✅ 0 B
Allocated bytes per simulated frame	< 4 KB	✅ 600 B
GC gen-0 count / 5-min stress run	< 600	✅ 273
GC gen-2 count / 5-min stress run	< 10	✅ 0
Managed heap growth over 5-min run	< 10 MB	✅ ~5.5 MB
Build passes after each fix	✅	✅
No new compiler warnings	✅	✅