Parallel Array¶
A multi-field record sequence represented by separate field arrays whose matching positions jointly form each logical record.
Core Idea¶
A parallel-array representation stores repeated records by field: each field has its own indexed array, and values at the same position jointly form one logical record. The shared positional identity matters more than any promised speedup.[ref-8a50be7c7a00][ref-1db0838ddb03]
Scope of Application¶
Intel's structure-of-arrays geometry example separates triangle-coordinate components for SIMD processing. Apache Arrow's record batches align same-length column arrays so one row position across them forms a logical row. These are two implementations of the same record-reconstruction pattern, not identical storage formats.[ref-8a50be7c7a00][ref-1db0838ddb03]
Clarity¶
With \(x[i]\), \(y[i]\) and \(m[i]\), record \(i\) is \((x[i],y[i],m[i])\). Sorting one field array alone would break the record association. An array of whole structures instead interleaves fields inside each record.[^ref-8a50be7c7a00]
Manages Complexity¶
Field-wise scans can avoid unused fields, while whole-record operations may gather from several arrays. The workload and hardware determine whether the trade-off is beneficial; the layout itself establishes only the separation and alignment.[^ref-3fd7898aa35d]
Abstract Reasoning¶
For field arrays \(F_1,\ldots,F_k\), define \(R_i=(F_1[i],\ldots,F_k[i])\). Coordinated reordering preserves the record mapping; uncoordinated reordering generally does not. Null masks and nested storage may add metadata without changing the logical same-position rule.[ref-4cec245deaea][ref-1db0838ddb03]
Knowledge Transfer¶
The triangle and table examples share logical records, separate field arrays and a common index. SIMD geometry processing and Arrow's typed columnar analytics are their different domain accents. Neither vectorization nor a particular memory-contiguity claim is constitutive.[ref-8a50be7c7a00][ref-4cec245deaea]
[^ref-8a50be7c7a00]: Intel, “Single Instruction Multiple Data Made Easy with Intel Implicit SPMD Program Compiler”, structure-of-arrays example. [^ref-4cec245deaea]: Apache Arrow, “Arrow Columnar Format”. [^ref-1db0838ddb03]: Apache Arrow, Format Glossary, record-batch and row definitions. [^ref-3fd7898aa35d]: Apache Arrow, Overview, columnar scan rationale.
Neighborhood in Abstraction Space¶
Parallel Array sits in a sparse region of the domain-specific corpus (65th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Storage & Lookup Data Structures (21 abstractions)
Nearest neighbors
- File Format — 0.86
- Rope (Data Structure) — 0.85
- Sorted Array — 0.85
- Column-Oriented Storage — 0.84
- Bloom Filter — 0.83
Computed from structural-signature embeddings · 2026-10-08