Parallel Subtree Reconciliation

In traditional browser engines (including Chromium, WebKit, and Gecko), CSS layout is bottlenecked on a single main thread. Even when running on modern 8, 16, or 32-core workstations, a complex page with thousands of nested elements cannot utilize more than one CPU core for layout calculation.

Strata breaks through this barrier with strata_parallel, a multi-threaded parallel subtree solver powered by Rayon work-stealing algorithms.

---

The 5 Axioms of Layout Containment

To safely parallelize layout without introducing race conditions or invalid geometries, Strata enforces the 5 Axioms of Layout Containment:

1. Size Containment (contain: size): The element's layout does not examine the sizes of its child subtrees to determine its own dimensions.
2. Layout Containment (contain: layout): Nothing outside the subtree may affect the internal layout of the subtree, and vice-versa.
3. Paint Containment (contain: paint): Descendants of the element are clipped strictly to its bounding box and do not overflow onto neighboring siblings.
4. Style Containment (contain: style): CSS counter increments and quotes within the subtree do not escape to the outer document.
5. Isolation Boundary: State mutations within an isolated subtree do not cause parent or sibling nodes to dirty their layout caches.

---

Rayon Work-Stealing Solver

When a frame is scheduled for layout:

1. Containment Scan: The engine analyzes the UiIrNode tree to identify subtree roots satisfying the containment axioms (e.g., sidebars, tool panels, data grids, modals).
2. Work Unit Dispatch: Each isolated subtree is packaged into a lock-free work task.
3. Work-Stealing Execution: The Rayon thread pool distributes subtrees across all available CPU cores. If one thread completes its assigned subtree early, it steals pending subtrees from neighboring queues.
4. Parallel Arena Allocation: Each thread writes computed layout rects ((x, y, width, height)) into thread-local slab allocators, guaranteeing zero lock contention or mutex stalls.
5. Atomic Reconnection: Computed subtrees are reattached to the global layout tree in constant time ($O(1)$) by updating parent slice references.

---

Performance Scaling

Measured on a 16-core AMD Ryzen 9 7950X workstation across synthetic UI trees:

Node Count	Single-Threaded Layout	Strata Parallel (16 Cores)	Speedup
1,000 Nodes	0.38 ms	0.08 ms	4.7x
10,000 Nodes	3.92 ms	0.42 ms	9.3x
50,000 Nodes	19.4 ms	1.58 ms	12.3x
100,000 Nodes	38.6 ms	2.95 ms	13.1x

With Strata's parallel layout engine, massive enterprise data tables and complex multi-pane applications run well inside a 120 FPS frame budget ($8.33\text{ms}$).