Notation & Abbreviation Reference¶
Every abbreviation used across this guide is listed here. On the rendered site, hovering over any underlined term in a document shows a tooltip with its full form and meaning — this page is the master reference you can bookmark or search.
How to Use¶
- Hover any dotted-underlined term on any page to see its full form in a tooltip.
- Search this page (Ctrl/Cmd + F) to look up a specific abbreviation.
- Abbreviations are grouped by domain. Each entry shows: Abbreviation → Full Form and a short explanation of how it is used in system design context.
Distributed Systems & Consistency¶
| Abbreviation | Full Form | Context |
|---|---|---|
| ACID | Atomicity, Consistency, Isolation, Durability | Properties that guarantee reliable database transactions |
| BASE | Basically Available, Soft state, Eventually consistent | Trade-off model for highly available distributed systems (contrast with ACID) |
| CAP | Consistency, Availability, Partition tolerance | Theorem: a distributed system can guarantee at most two of these three properties simultaneously |
| CP | Consistency and Partition tolerance | A CAP trade-off: system chooses correctness over availability during partitions |
| AP | Availability and Partition tolerance | A CAP trade-off: system stays available but may return stale data during partitions |
| CRDT | Conflict-free Replicated Data Type | Data structure that can be merged across replicas without conflicts (e.g., G-Counter, LWW-Register) |
| LWW | Last-Write-Wins | Conflict resolution strategy that keeps the most recent write (by timestamp) |
| HLC | Hybrid Logical Clock | Clock combining physical time and logical counters for distributed event ordering |
| 2PC | Two-Phase Commit | Protocol ensuring all-or-nothing writes across multiple nodes (prepare → commit/abort) |
| CQRS | Command Query Responsibility Segregation | Pattern separating write (command) and read (query) models |
| DAG | Directed Acyclic Graph | Graph with directed edges and no cycles; used to model task dependencies (e.g. workflow DAGs) |
| ZAB | Zookeeper Atomic Broadcast | Consensus and leader-election protocol used by Apache ZooKeeper |
| OT | Operational Transformation | Algorithm for real-time collaborative editing that merges concurrent edits without conflicts |
Storage & Databases¶
| Abbreviation | Full Form | Context |
|---|---|---|
| SQL | Structured Query Language | Language for relational databases; also used as a shorthand for "relational database" |
| OLTP | Online Transaction Processing | Workload optimised for short reads/writes (e.g. web apps, payment systems) |
| OLAP | Online Analytical Processing | Workload optimised for large analytical scans (e.g. data warehouses, BI) |
| LSM | Log-Structured Merge-tree | Write-optimised storage engine (used by Cassandra, RocksDB) — appends to memtable then flushes to SSTables |
| SSTable | Sorted String Table | Immutable on-disk file produced by an LSM flush; keys are sorted for efficient range scans |
| B+tree | Balanced Plus Tree | Disk-optimised search tree used by most RDBMS (PostgreSQL, MySQL) for index pages |
| WAL | Write-Ahead Log | Durability mechanism: changes written to log before being applied to main storage |
| CDC | Change Data Capture | Technique for streaming row-level changes out of a database (e.g. Debezium over Postgres WAL) |
| STCS | Size-Tiered Compaction Strategy | Cassandra compaction: merges similarly-sized SSTables; good for write-heavy workloads |
| LCS | Leveled Compaction Strategy | Cassandra compaction: organises SSTables into fixed-size levels; better for read-heavy workloads |
| TWCS | Time-Window Compaction Strategy | Cassandra compaction for time-series data; compacts within rolling time windows |
| KV | Key-Value | Simplest data model: store and retrieve opaque values by key (e.g. Redis, DynamoDB) |
| FK | Foreign Key | Relational constraint linking a column to a primary key in another table |
| RDD | Resilient Distributed Dataset | Spark's core abstraction: an immutable, fault-tolerant distributed collection |
Networking & Protocols¶
| Abbreviation | Full Form | Context |
|---|---|---|
| HTTP | Hypertext Transfer Protocol | Application-layer protocol for web communication |
| HTTPS | Hypertext Transfer Protocol Secure | HTTP over TLS — encrypted web communication |
| TCP | Transmission Control Protocol | Reliable, ordered, connection-oriented transport protocol |
| UDP | User Datagram Protocol | Connectionless, best-effort transport; lower latency than TCP (used in video, DNS) |
| TLS | Transport Layer Security | Cryptographic protocol for secure communication; successor to SSL |
| SSL | Secure Sockets Layer | Predecessor to TLS; deprecated but term still used colloquially |
| mTLS | Mutual TLS | Both client and server authenticate each other with certificates; used in service mesh (e.g. Istio) |
| DNS | Domain Name System | Resolves human-readable hostnames to IP addresses |
| IP | Internet Protocol | Network-layer protocol for addressing and routing packets |
| gRPC | Google Remote Procedure Call | High-performance RPC framework using HTTP/2 and Protocol Buffers |
| RPC | Remote Procedure Call | Pattern for calling functions on a remote server as if local |
| REST | Representational State Transfer | Architectural style for HTTP APIs using resource-based URLs and standard methods |
| WebRTC | Web Real-Time Communication | Browser API for peer-to-peer audio, video, and data channels |
| WSS | WebSocket Secure | WebSocket protocol over TLS |
| SSE | Server-Sent Events | HTTP-based one-way streaming from server to browser |
| RTT | Round-Trip Time | Time for a message to travel from sender to receiver and back |
| FQDN | Fully Qualified Domain Name | Complete domain name including all labels (e.g. api.example.com) |
| AMQP | Advanced Message Queuing Protocol | Wire-level messaging protocol; used by RabbitMQ |
| NTP | Network Time Protocol | Protocol for synchronising clocks across networked computers |
Performance & Reliability¶
| Abbreviation | Full Form | Context |
|---|---|---|
| SLA | Service Level Agreement | Contractual commitment between provider and customer (e.g. 99.9% uptime) |
| SLO | Service Level Objective | Internal target derived from an SLA (e.g. P99 latency < 200 ms) |
| SLI | Service Level Indicator | Metric used to measure whether an SLO is being met (e.g. request success rate) |
| P50 | 50th Percentile | Median — 50% of requests are faster than this |
| P95 | 95th Percentile | 95% of requests are faster than this; filters out top-5% outliers |
| P99 | 99th Percentile | 99% of requests are faster than this; the standard "tail latency" benchmark |
| QPS | Queries Per Second | Request throughput — often used for read-heavy systems |
| RPS | Requests Per Second | Request throughput — general alternative to QPS |
| RPM | Revenue Per Mille / Requests Per Minute | In ads: revenue per 1,000 impressions. In infra: requests per minute |
| IOPS | Input/Output Operations Per Second | Disk or storage throughput metric |
| MTBF | Mean Time Between Failures | Average time between successive failures of a system |
| MTTR | Mean Time To Recovery | Average time to restore a system after a failure |
| RPO | Recovery Point Objective | Maximum acceptable data loss measured in time (e.g. 5 minutes of data) |
| RTO | Recovery Time Objective | Maximum acceptable downtime before service must be restored |
| SPOF | Single Point of Failure | A component whose failure brings down the entire system |
| NFR | Non-Functional Requirement | System quality attribute: latency, availability, scalability, security, etc. |
| QoS | Quality of Service | Mechanisms for prioritising traffic or guaranteeing performance levels |
Caching & Storage Hardware¶
| Abbreviation | Full Form | Context |
|---|---|---|
| TTL | Time To Live | Duration after which a cached entry or DNS record expires |
| SSD | Solid State Drive | Flash-based storage; much faster random I/O than HDD |
| HDD | Hard Disk Drive | Magnetic spinning disk; high capacity, low random IOPS |
| RAM | Random Access Memory | Fast volatile memory; used for in-memory caches (Redis, Memcached) |
| GB | Gigabyte | 10⁹ bytes (or 2³⁰ bytes in binary) |
| MB | Megabyte | 10⁶ bytes |
| KB | Kilobyte | 10³ bytes |
| TB | Terabyte | 10¹² bytes |
| PB | Petabyte | 10¹⁵ bytes |
| I/O | Input/Output | Read/write operations to storage or peripherals |
| OOM | Out of Memory | Condition where a process exceeds available RAM; often triggers eviction or crash |
| GC | Garbage Collection | Automatic memory reclamation; GC pauses can cause latency spikes |
Infrastructure & Cloud¶
| Abbreviation | Full Form | Context |
|---|---|---|
| API | Application Programming Interface | Contract defining how two software components communicate |
| SDK | Software Development Kit | Library + tools for interacting with a platform or service |
| CI/CD | Continuous Integration / Continuous Deployment | Automated pipeline: build → test → deploy on every commit |
| CDN | Content Delivery Network | Geographically distributed cache for static assets and media |
| DNS | Domain Name System | See Networking section |
| VPC | Virtual Private Cloud | Isolated private network inside a public cloud provider |
| AZ | Availability Zone | Isolated data-centre within a cloud region; failure domains |
| CRR | Cross-Region Replication | Automatically replicating data to a second geographic region |
| S3 | Simple Storage Service | AWS object storage service; often used generically for object/blob storage |
| GCS | Google Cloud Storage | Google Cloud's object storage service |
| LB | Load Balancer | Distributes incoming traffic across multiple backend instances |
| HPA | Horizontal Pod Autoscaler | Kubernetes controller that scales pod replicas based on CPU/memory metrics |
| VPA | Vertical Pod Autoscaler | Kubernetes controller that adjusts pod CPU/memory requests automatically |
| SRE | Site Reliability Engineering | Discipline applying software engineering to operations (Google-origin) |
| DevOps | Development and Operations | Culture and practice of shared ownership across dev and ops teams |
| MLOps | Machine Learning Operations | DevOps practices applied to ML: versioning, monitoring, retraining pipelines |
| SaaS | Software as a Service | Cloud delivery model: vendor manages everything; customer uses via browser |
| DLQ | Dead Letter Queue | Queue that receives messages that could not be processed after retries |
| MPSC | Multi-Producer Single-Consumer | Queue pattern common in high-throughput logging pipelines |
| P2P | Peer-to-Peer | Network architecture where nodes communicate directly (no central server) |
| UUID | Universally Unique Identifier | 128-bit identifier generated to be unique without central coordination |
| DDoS | Distributed Denial of Service | Attack using many sources to overwhelm a system with traffic |
Security & Privacy¶
| Abbreviation | Full Form | Context |
|---|---|---|
| TLS | Transport Layer Security | See Networking section |
| JWT | JSON Web Token | Compact, signed token for stateless authentication and claims |
| RBAC | Role-Based Access Control | Permissions granted to roles; users inherit role permissions |
| ACL | Access Control List | Per-resource permission list mapping principals to allowed actions |
| GDPR | General Data Protection Regulation | EU regulation governing personal data collection, storage, and consent |
| PII | Personally Identifiable Information | Data that can identify an individual (name, email, IP address, etc.) |
| XSS | Cross-Site Scripting | Web vulnerability injecting malicious scripts into pages viewed by other users |
| PCI-DSS | Payment Card Industry Data Security Standard | Compliance framework for systems handling card payment data |
| PCI | Payment Card Industry | Industry body that sets PCI-DSS standards |
| PAN | Primary Account Number | The 16-digit card number; must be tokenised or encrypted, never stored in plain text |
| SEO | Search Engine Optimization | Practices to improve a page's ranking in organic search results |
Payments & Finance¶
| Abbreviation | Full Form | Context |
|---|---|---|
| PSP | Payment Service Provider | Third-party that processes card/bank payments (e.g. Stripe, Adyen, PayPal) |
| ACH | Automated Clearing House | US electronic bank-to-bank transfer network (direct debit / payroll) |
| SEPA | Single Euro Payments Area | EU equivalent of ACH for euro-denominated bank transfers |
ML & AI Fundamentals¶
| Abbreviation | Full Form | Context |
|---|---|---|
| LLM | Large Language Model | Transformer-based model trained on text at scale (GPT-4, Gemini, Claude) |
| RAG | Retrieval-Augmented Generation | Architecture combining a retriever with an LLM to ground answers in documents |
| MLP | Multi-Layer Perceptron | Fully-connected feedforward neural network |
| CNN | Convolutional Neural Network | Network using convolutional filters; standard for image and audio tasks |
| RNN | Recurrent Neural Network | Sequential model with hidden state; largely superseded by Transformers |
| LSTM | Long Short-Term Memory | RNN variant with gating that mitigates the vanishing gradient problem |
| BERT | Bidirectional Encoder Representations from Transformers | Pre-trained encoder model; fine-tuned for NLU tasks |
| ViT | Vision Transformer | Transformer applied to image patches; state-of-the-art for image understanding |
| GNN | Graph Neural Network | Neural network that operates on graph-structured data |
| MoE | Mixture of Experts | Architecture with many "expert" sub-networks; only a few are active per token (e.g. Mixtral) |
| CLIP | Contrastive Language-Image Pre-Training | OpenAI model that jointly embeds images and text; foundational for multimodal retrieval |
| ANN | Approximate Nearest Neighbor | Fast similarity search that trades exact results for speed (HNSW, FAISS, ScaNN) |
| KNN | K-Nearest Neighbors | Exact nearest-neighbor search; practical only for small datasets |
| HNSW | Hierarchical Navigable Small World | Graph-based ANN index algorithm; best recall/speed trade-off for dense vectors |
| FAISS | Facebook AI Similarity Search | Meta's library of ANN indexes; GPU-accelerated options available |
| BM25 | Best Matching 25 | Probabilistic ranking function for keyword (sparse) retrieval |
| TF-IDF | Term Frequency-Inverse Document Frequency | Classical keyword weighting: high weight for rare terms in a document |
| IDF | Inverse Document Frequency | The log-scaled penalty for common terms; the "IDF" component of TF-IDF |
| NLP | Natural Language Processing | Broad field of ML applied to text understanding and generation |
| NLU | Natural Language Understanding | Subset of NLP focused on intent, entities, and semantic meaning |
| NER | Named Entity Recognition | Task of labelling entities (person, location, organisation) in text |
| NLI | Natural Language Inference | Task of determining if a premise entails, contradicts, or is neutral to a hypothesis |
| OCR | Optical Character Recognition | Converting images of text into machine-readable characters |
ML Training¶
| Abbreviation | Full Form | Context |
|---|---|---|
| RLHF | Reinforcement Learning from Human Feedback | Training paradigm: LLM fine-tuned with a reward model trained from human preferences |
| SFT | Supervised Fine-Tuning | First stage of RLHF: fine-tune a base LLM on curated instruction-response pairs |
| DPO | Direct Preference Optimization | RLHF alternative: optimises directly on preference pairs without a separate reward model |
| PPO | Proximal Policy Optimization | RL algorithm used in RLHF reward phase to constrain policy updates |
| LoRA | Low-Rank Adaptation | PEFT technique: inject small trainable rank-decomposition matrices into frozen weights |
| QLoRA | Quantized Low-Rank Adaptation | LoRA + 4-bit quantized base weights; enables fine-tuning 65B+ models on a single GPU |
| PEFT | Parameter-Efficient Fine-Tuning | Family of techniques that update only a small fraction of weights (LoRA, adapters, prefix tuning) |
| FSDP | Fully Sharded Data Parallel | PyTorch training strategy: shards model parameters, gradients, and optimizer state across GPUs |
| DDP | Distributed Data Parallel | PyTorch training strategy: each GPU holds a full model copy; gradients are all-reduced |
| ZeRO | Zero Redundancy Optimizer | DeepSpeed memory optimization: eliminates redundant parameter/gradient/optimizer state across GPUs |
| FP16 | 16-bit Floating Point | Half-precision format; reduces memory by 2× vs FP32; may require loss scaling |
| BF16 | BFloat16 | 16-bit format with the same exponent range as FP32; more numerically stable than FP16 for training |
| FP32 | 32-bit Floating Point | Single-precision; standard for master weights and optimizer state |
| INT8 | 8-bit Integer | Quantization format for inference; reduces memory and latency at the cost of precision |
ML Serving & Infrastructure¶
| Abbreviation | Full Form | Context |
|---|---|---|
| GPU | Graphics Processing Unit | Massively parallel processor; dominant compute for neural network training and inference |
| TPU | Tensor Processing Unit | Google's custom ASIC for matrix multiplications; used in Google Cloud |
| CPU | Central Processing Unit | General-purpose processor; used for preprocessing, light inference, orchestration |
| A100 | NVIDIA A100 GPU | 80 GB HBM2e; dominant for LLM training and large-batch inference |
| H100 | NVIDIA H100 GPU | Successor to A100; Transformer Engine, NVLink 4; 3× the LLM training throughput of A100 |
| TPM | Tokens Per Minute | Rate limit unit for LLM APIs |
| DAG | Directed Acyclic Graph | See Distributed Systems section; also used for ML pipeline and workflow graphs |
| ETL | Extract, Transform, Load | Classic data pipeline pattern: pull from source, transform, load into warehouse |
| ELT | Extract, Load, Transform | Modern variant: load raw data first, transform inside the data warehouse |
ML Metrics & Evaluation¶
| Abbreviation | Full Form | Context |
|---|---|---|
| AUC | Area Under the ROC Curve | Model discrimination metric: probability that a random positive ranks above a random negative |
| ECE | Expected Calibration Error | Measures calibration: weighted average gap between predicted confidence and actual accuracy |
| BLEU | Bilingual Evaluation Understudy | N-gram precision metric for machine translation quality |
| ROUGE | Recall-Oriented Understudy for Gisting Evaluation | Recall-based metric for summarisation; ROUGE-L measures longest common subsequence |
| WER | Word Error Rate | Speech recognition metric: (substitutions + deletions + insertions) / total reference words |
| CTC | Connectionist Temporal Classification | Loss function for sequence alignment without frame-level labels; used in ASR and OCR |
| ASR | Automatic Speech Recognition | Task of converting spoken audio to text (e.g. Whisper, DeepSpeech) |
| TTS | Text-to-Speech | Task of synthesising spoken audio from text |
Ads & Monetisation¶
| Abbreviation | Full Form | Context |
|---|---|---|
| CTR | Click-Through Rate | Fraction of impressions that result in a click; core signal for ad quality |
| pCTR | Predicted Click-Through Rate | Model's estimate of P(click | impression, context); used in auction ranking |
| CVR | Conversion Rate | Fraction of clicks that result in a conversion (purchase, sign-up, install) |
| pCVR | Predicted Conversion Rate | Model's estimate of P(conversion | click, context) |
| CPC | Cost Per Click | Billing model: advertiser pays per click on their ad |
| CPM | Cost Per Mille | Billing model: advertiser pays per 1,000 impressions |
| CPA | Cost Per Action | Billing model: advertiser pays per conversion/acquisition |
| eCPM | Effective Cost Per Mille | Expected revenue per 1,000 impressions; used to normalise bids across billing models |
| RPM | Revenue Per Mille | Publisher-side revenue per 1,000 impressions |
| ROAS | Return on Ad Spend | Advertiser metric: conversion value earned per dollar spent on ads |
| tCPA | Target Cost Per Acquisition | Automated bidding objective: platform adjusts bids to achieve a specified CPA target |
| tROAS | Target Return on Ad Spend | Automated bidding objective: platform adjusts bids to achieve a specified ROAS target |
| GSP | Generalized Second-Price | Multi-slot auction mechanism used in sponsored search; winners pay the next competitor's externality |
| VCG | Vickrey-Clarke-Groves | Theoretically efficient and incentive-compatible auction mechanism; charges each winner their externality on others |
| IVT | Invalid Traffic | Non-human or fraudulent ad traffic (bots, click farms); excluded from billing and training labels |
| LTV | Lifetime Value | Expected total revenue from a user (or customer) over their relationship with the product |
| IPS | Inverse Propensity Scoring | Debiasing technique: reweights logged examples by the inverse probability of their being observed |
Algorithms & Consensus¶
| Abbreviation | Full Form | Context |
|---|---|---|
| FIFO | First In, First Out | Queue ordering: oldest item is processed first |
| LWW | Last-Write-Wins | See Distributed Systems section |
| CRDT | Conflict-free Replicated Data Type | See Distributed Systems section |
| ZAB | Zookeeper Atomic Broadcast | See Distributed Systems section |
| PID | Proportional-Integral-Derivative | Classic control loop algorithm; used in auto-bidding and pacing controllers |
| BM25 | Best Matching 25 | See ML Fundamentals section |
Abbreviation not listed here?
Open an issue or PR on the GitHub repository. All additions are welcome.