glossary
Every technical term the site relies on, defined once. Entries are reviewed by hand (SPEC §27).
no terms match.
A
- ANN search Approximate Nearest Neighbor search — algorithms that return near-best matches from a vector set while trading recall for latency and memory; the backbone of modern semantic search and RAG. application · memory
- AVX2 Intel's 256-bit SIMD instruction set extension, the practical sweet spot for SIMD on modern x86 before AVX-512 frequency and availability issues bite. cpu
B
- blk-mq Linux's multi-queue block I/O layer, replacing the legacy single-queue path; uses per-CPU software queues and per-device hardware queues to scale to modern NVMe parallelism. kernel · storage
- branch predictor CPU hardware that speculates the outcome of upcoming branches so the pipeline can keep fetching instructions without waiting for the condition; mispredictions cost a full pipeline flush. cpu
C
- cache line The unit of data transfer between a CPU's caches and main memory; 64 bytes on x86_64 (and most modern Arm), 128 bytes on Apple M-series L2. The atomic unit of coherence traffic. cpu · cache
- context switch The act of saving the CPU state of one thread (or process) and restoring another's; involves kernel work, TLB effects, and cache pollution that show up as latency tail outliers. kernel · cpu
- copy-on-write Memory-management technique that shares a read-only mapping until a write triggers a private copy; the reason fork() is cheap and why a mis-tuned workload can take a latency spike on the first store. kernel · memory
D
E
F
H
- HNSW Hierarchical Navigable Small World — the dominant graph-based approximate nearest neighbor algorithm in modern vector databases, used by FAISS, hnswlib, Qdrant, Weaviate, Milvus, and pgvector. application · memory · cache
- huge page A memory page larger than the architecture's base size (2 MB or 1 GB on x86_64 vs 4 KB baseline) that lets a single TLB entry cover more memory, reducing page-table walks for large working sets. memory · kernel
I
- I/O depth The number of I/O requests in flight at once against a device or interface; the primary knob that trades latency for throughput on block storage and async networking. storage · kernel
- io_uring Linux async I/O interface (kernel 5.1+) built on two shared-memory ring buffers that batches syscalls and amortizes kernel crossings for low-latency file, network, and storage work. syscall · kernel
M
- madvise Linux syscall that hints the kernel about a memory region's future access pattern (sequential, random, will-need, don't-need, hugepage, etc.); the main tuning dial for mmap-backed workloads. syscall · memory
- memory bandwidth Sustained rate at which a CPU can read from or write to main memory — measured in GB/s per socket — and one of the two fundamental ceilings (with latency) for data-intensive workloads. memory · cpu · bus / dma
- memory mapping Mapping a file or anonymous region into a process's virtual address space via mmap(2) so that loads and stores translate directly into page-cache access or demand-paged backing storage. syscall · memory · kernel
- MESI protocol The dominant cache-coherence protocol in modern multicore CPUs, tracking each cache line in one of four states — Modified, Exclusive, Shared, Invalid — to keep cores' views of memory consistent. cpu · cache
N
P
- page cache Linux kernel's in-memory cache of filesystem page contents; the reason repeat reads of a warm file hit DRAM speed and why O_DIRECT exists for workloads that want to bypass it. kernel · memory · storage
- PCIe TLP Transaction Layer Packet — the unit of communication over a PCI Express link, carrying memory reads/writes, completions, and messages between CPU root complex and devices. bus / dma · storage
- prefetcher CPU hardware that predicts upcoming memory accesses and pulls cache lines into L1/L2 before the demand load arrives; effective on regular stride patterns, useless on pointer chasing. cpu · cache
Q
R
S
- SIMD Single Instruction, Multiple Data — CPU instructions that perform the same operation on a vector of values in parallel; SSE, AVX, AVX2, AVX-512 on x86, NEON/SVE on Arm. cpu
- syscall A controlled transition from user-mode code into the kernel to request a privileged service (I/O, memory, scheduling); the per-crossing cost is the floor below which many performance tricks aim to push. syscall · kernel
T
- TLB Translation Lookaside Buffer — a small on-CPU cache of recent virtual-to-physical page-table translations; a miss triggers a multi-level page walk that is one of the quiet latency costs on large working sets. cpu · memory
- TLB shootdown The kernel's mechanism for invalidating stale TLB entries on other CPUs after a page table change, implemented via an inter-processor interrupt that stalls every target core for the duration. kernel · cpu · memory
V
- vDSO Virtual dynamic shared object — a tiny kernel-provided shared library mapped into every process that lets common read-only syscalls (clock_gettime, getpid) run without crossing into the kernel. syscall · kernel
- vector quantization Compressing high-dimensional vectors into shorter codes (product, scalar, or binary) so that ANN indexes fit in cache or memory; trades a small recall cost for a large memory reduction. memory · application