Research area · programme
Distributed Systems
The database floor: replication, LSM engines, grey failures, fleet upgrades under measurement. 76 papers in the vault, 23 drawn here.
23 papers drawn here · 78 curated relations touch the area · 4 authored papers. The graph opens straight on this view; the map and every other area stay one click away.
Loading the full graph…
Papers drawn in Distributed Systems (23)
Every paper drawn in this area, sorted by curated degree. Click a title to open its ego graph on the canvas above, or follow the DOI / arXiv link to the publisher record.
| Paper | Authors | Year | Identifier / Venue | Status | Relations |
|---|---|---|---|---|---|
| What The Log: Compile-Time AST Template Mining and Two-Tier Statistical Temporal Compression for Agent-Queryable Parquet Log Lakehouses Extracts 53,848 log templates directly from polyglot source-code ASTs at compile time and compresses consecutive log bursts via two-tier statistical temporal run-length encoding into agent-queryable Apache Parquet. | Vidigal | 2026 | Yugabyte, Inc. | vault draft | 10 |
| Machine-Checked Log Compression: Zero-Axiom Lean 4 Verification and C-ABI Differential Testing for Streaming Log Parsers and Temporal RLE Proves round-trip multiset conservation and temporal run-length encoding invariants for streaming log parsers in Lean 4 with zero axioms, paired with C-ABI differential fuzzing. | Vidigal | 2026 | Yugabyte, Inc. | vault draft | 4 |
| Most Health Rules Do Not Know Their Fleet: Population-Calibrated Thresholds, Three-Layer Effective Defaults, and Stratified Epidemiology of 212 Database Diagnostic Rules Evaluates 212 database diagnostic health rules across production telemetry, replacing static vendor thresholds with population-calibrated quantiles and stratified prevalence priors. | Vidigal | 2026 | Yugabyte, Inc. | vault draft | 2 |
| Pre-Registering the Speedup: Cost-Bounded, Multiset-Equivalent Performance Engineering of a Terabyte-Scale Log-to-Parquet Engine Shows how streaming gzip/tar ingestion, right-sized Parquet bloom filters (v0.4.20 cutting silver Parquet bytes by ~0.5x), and bounded categorical ArgStats (v0.4.21 reducing peak RSS from 16-22 GB to 2-6 GB on YBA application logs) prevent OOM kills and quadratic IntervalSet stalls on 72-96.73 GiB support bundles. | Vidigal | 2026 | Yugabyte, Inc. | vault draft | 1 |
| The log-structured merge-tree (LSM-tree) LSM read cost = seeks across levels plus nexts over obsolete versions/tombstones. For YugabyteDB DocDB (RocksDB-derived) this explains why EXPLAIN (ANALYZE, DIST, DEBUG) rocksdb_number_db_seek/next and pg_stat_statements docdb_obsolete_rows_scanned diverge from rows returned after bulk deletes; | O’Neil et al. | 1996 | DOI:10.1007/s002360050048 | peer-reviewed (vault claim) | 15 |
| SILK+ Preventing Latency Spikes in Log-Structured Merge Key-Value Stores Running Heterogeneous Workloads SILK / SILK+ demonstrates that p99 tail latency in RocksDB spikes by 2-3 orders of magnitude (from sub-ms to ~1s) under sustained writes due to I/O bandwidth competition between foreground flushes/L0-L1 compactions and deep L2+ compactions (which trigger memtable/L0 write stalls). | Balmau et al. | 2018 | DOI:10.1145/3380905 | peer-reviewed (vault claim) | 8 |
| Spanner: Google's Globally-Distributed Database Spanner's leader reads, TrueTime-bounded uncertainty and bounded-staleness snapshot reads map to YugabyteDB tablet-leader reads, max_clock_skew-driven read restarts (40001), and yb_read_from_followers + yb_follower_read_staleness_ms. Profiler implication: cross-region latency is a leader-placement problem first; | Corbett et al. | 2012 | Proceedings of the 10th USENIX Symposium on Operating Systems Design and Implementation (OSDI '12) | identity-only record | 7 |
| CockroachDB: The Resilient Geo-Distributed SQL Database CockroachDB is the closest public comparator to YSQL; useful contrast: CRDB range-partitions by default whereas YSQL hash-shards the first PK column by default (yb_use_hash_splitting_by_default), which changes which predicates become multi-tablet fan-out scans. | Taft et al. | 2020 | DOI:10.1145/3318464.3386134 | peer-reviewed (vault claim) | 6 |
| RocksDB: Evolution of Development Priorities in a Key-Value Store Serving Large-Scale Applications. DocDB is a RocksDB fork; RocksDB's evolution toward space efficiency, compaction tuning and write-stall control explains YB tserver CPU/IO hot spots (compaction, flush, block cache) and YB's own SST-file-count throttling (sst_files_soft_limit 24/hard 48) that replaces RocksDB L0 stall triggers. | Dong & Stumm | 2021 | DOI:10.1145/3483840 | peer-reviewed (vault claim) | 6 |
| Realizing Strong Determinism Contract on Log-Structured Merge Key-Value Stores Vigil-KV co-designs LSM compaction/flush scheduling with NVMe Predictable Latency Mode (PLM) windows, cutting tail latency by 3.19x and average latency by 34% across Facebook and Yahoo workloads. | Kwon et al. | 2023 | DOI:10.1145/3582695 | peer-reviewed (vault claim) | 5 |
| TiDB: A Raft-based HTAP Database TiDB's Placement Driver balances region replicas, leaders and hot regions from store heartbeats as separate objectives. Supports treating YB leader skew and byte skew as two distinct per-universe metrics; hot-tablet detection needs load (ops/s), which tablet reports do not carry. | Huang et al. | 2020 | DOI:10.14778/3415478.3415535 | peer-reviewed (vault claim) | 5 |
| WiscKey WiscKey separates keys (stored sorted in the LSM-tree) from values (stored in an append-only vLog), proving that LSM compaction re-sorts values unnecessarily when only keys are needed for indexing. Reduces write amplification by an order of magnitude and achieves 2.5x-111x faster DB load and 1.6x-14x faster YCSB workloads than LevelDB. | Lu et al. | 2017 | DOI:10.1145/3033273 | peer-reviewed (vault claim) | 4 |
| Calvin Calvin avoids distributed commit cost by deterministic pre-ordering; contrasts with YugabyteDB wait-queue concurrency where contention surfaces as 40001/40P01 retries and ConflictResolution_WaitOnConflictingTxns ASH samples. Background for treating hot-row contention as a schema problem, not a tuning knob. | Thomson et al. | 2012 | DOI:10.1145/2213836.2213838 | peer-reviewed (vault claim) | 3 |
| Framework for SQL Error Message Design: A Data-Driven Approach Data-driven framework of nine SQL error-message design guidelines, derived from 128 messages across eight DBMSs, 4,796 queries from 311 participants and content analysis of 1,505 user suggestions. | Taipalus & Grahn | 2023 | DOI:10.1145/3607180 | peer-reviewed (vault claim) | 3 |
| Zero-Downtime SQL Database Schema Evolution for Continuous Deployment | Jong et al. | 2017 | DOI:10.1109/ICSE-SEIP.2017.5 | peer-reviewed (vault claim) | 3 |
| A Critique of ANSI SQL Isolation Levels. Demonstrates that the ANSI SQL-92 phenomenon definitions (dirty read, non-repeatable read, phantom) are ambiguous, formalizes Snapshot Isolation (SI), and characterizes write-skew anomalies. | Berenson & O'Neil | 1995 | DOI:10.1145/223784.223785 | peer-reviewed (vault claim) | 2 |
| Access path selection in a relational database management system Origin of cost-based access path selection; YugabyteDB's CBO (yb_enable_cbo=on) extends it with network round-trip and LSM seek/next costs, so plan choice is still only as good as statistics, now with RPC count as a first-class cost. | Selinger et al. | 1979 | DOI:10.1145/582095.582099 | peer-reviewed (vault claim) | 2 |
| Similarity Metrics for SQL Query Clustering Compares similarity metrics for clustering SQL query logs and shows structural feature extraction yields usable workload clusters. Operational use for PM workload-shape questions from support bundles: cluster pgaudit/slow-query statement fingerprints (artifact.query_event; | Kul et al. | 2018 | DOI:10.1109/TKDE.2018.2831214 | peer-reviewed (vault claim) | 2 |
| Spanner | Bacon et al. | 2017 | DOI:10.1145/3035918.3056103 | peer-reviewed (vault claim) | 2 |
| Towards Robust Distributed Systems. Brewer, E. A. (2000). "`Towards Robust Distributed Systems.`" _Keynote at the 19th Annual ACM Symposium on Principles of Distributed Computing (PODC)_; formalized by Gilbert, S. and Lynch, N. (2002), "`Brewer's Conjecture and the Feasibility of Consistent, Available, Partition-Tolerant Web Services,`" _ACM SIGACT News_, 33(2), 51–59. | Brewer | 2000 | DOI:10.1145/564585.564601 | peer-reviewed (vault claim) | 2 |
| WAL Failover (CockroachDB documentation) CockroachDB treats disk stalls by magnitude: a stalled store blocks writes for its leases until the stall clears or the node is crashed, by default after 20 s (max sync duration); WAL failover lets it ride out ~40 s. A single-stall-length threshold, not a volume budget, is the vendor's failover trigger. | Cockroach Labs | 2026 | — | identity-only record | 2 |
| Load-Based Splitting (CockroachDB documentation) | Cockroach Labs | 2026 | — | identity-only record | 1 |
| Serializable Snapshot Isolation in PostgreSQL PostgreSQL's SSI implementation (SIREAD locks, safe snapshots, read-only optimisations) and its measured false-positive abort rate. YSQL inherits PG error semantics (SQLSTATE 40001), so error surveillance must separate SSI/conflict aborts from read-restart errors, which YB raises under the same code. | Ports & Grittner | 2012 | arXiv:1208.4179 | preprint | 1 |
Curated semantic relation notes touching Distributed Systems (40)
extends → Realizing Strong Determinism Contract on Log-Structured Merge Key-Value Stores → SILK+ Preventing Latency Spikes in Log-Structured Merge Key-Value Stores Running Heterogeneous Workloads
Co-designs LSM compaction scheduling with NVMe PLM deterministic windows beyond software-only SILK
Drafted by: agent draft (gemini)
extends → How good are query optimizers, really? → Access path selection in a relational database management system
Empirically evaluates Selinger-style cost-based optimization; cardinality estimation dominates.
Drafted by: agent draft (gemini)
complements ↔ KVell ↔ SILK+ Preventing Latency Spikes in Log-Structured Merge Key-Value Stores Running Heterogeneous Workloads
Shows how NVMe parallelism shifts KV store bottlenecks from disk bandwidth to CPU/memory-copy and compaction interference
Drafted by: agent draft (gemini)
complements ↔ WiscKey ↔ RocksDB: Evolution of Development Priorities in a Key-Value Store Serving Large-Scale Applications.
Key-value separation (WiscKey / BlobDB) reduces LSM compaction WAF in RocksDB
Drafted by: agent draft (gemini)
extends → Spanner → Spanner: Google's Globally-Distributed Database
Spanner SQL system incl. schema change at scale
Drafted by: agent draft (claude)
extends → RocksDB: Evolution of Development Priorities in a Key-Value Store Serving Large-Scale Applications. → SILK+ Preventing Latency Spikes in Log-Structured Merge Key-Value Stores Running Heterogeneous Workloads
Eliminates RocksDB p99 latency spikes caused by L0/memtable write stalls via compaction I/O scheduling and preemption
Drafted by: agent draft (gemini)
complements ↔ Neuro-Symbolic Verification on Instruction Following of LLMs ↔ Machine-Checked Log Compression: Zero-Axiom Lean 4 Verification and C-ABI Differential Testing for Streaming Log Parsers and Temporal RLE
Complements machine-checked Lean 4 parser verification with neuro-symbolic constraint verification for instruction-following agents.
Drafted by: agent draft (gemini)
operationalizes → Serializable Snapshot Isolation in PostgreSQL → Serializable isolation for snapshot databases
PostgreSQL implementation of SSI
Drafted by: agent draft (gemini)
contrasts with ↔ PebblesDB ↔ WiscKey
Reduces LSM WAF via guard-based Fragmented LSM (FLSM) without separating values into a vLog
Drafted by: agent draft (gemini)
complements ↔ "What makes my queries slow?": Subgroup Discovery for SQL Workload Analysis ↔ Similarity Metrics for SQL Query Clustering
Outcome-aware subgroup discovery vs unsupervised SQL clustering
Drafted by: agent draft (claude)
complements ↔ CockroachDB: The Resilient Geo-Distributed SQL Database ↔ Spanner: Google's Globally-Distributed Database
CockroachDB adapts Spanner design with HLC instead of TrueTime.
Drafted by: agent draft (gemini)
extends → CockroachDB: The Resilient Geo-Distributed SQL Database → Spanner: Google's Globally-Distributed Database
Adapts Spanner's distributed transactional architecture to commodity clocks and Raft/LSM (RocksDB/Pebble)
Drafted by: agent draft (gemini)
extends → Framework for SQL Error Message Design: A Data-Driven Approach → Compiler Error Messages Considered Unhelpful
Carries general error-message guideline research into the SQL/DBMS domain.
Drafted by: agent draft (gemini)
extends → Framework for SQL Error Message Design: A Data-Driven Approach → Error messages in relational database management systems: A comparison of effectiveness, usefulness, and user confidence
From comparing four DBMS message sets to deriving design guidelines.
Drafted by: agent draft (gemini)
operationalizes → Enhanced SQL error messages facilitate faster error fixing → Framework for SQL Error Message Design: A Data-Driven Approach
Experimental test of the nine-guideline framework: enhanced messages faster to fix.
Drafted by: agent draft (gemini)
contrasts with ↔ Calvin ↔ Spanner: Google's Globally-Distributed Database
Deterministic pre-ordering vs 2PC+TrueTime.
Drafted by: agent draft (gemini)
complements ↔ Most Health Rules Do Not Know Their Fleet: Population-Calibrated Thresholds, Three-Layer Effective Defaults, and Stratified Epidemiology of 212 Database Diagnostic Rules ↔ ExplainIt! -- A Declarative Root-cause Analysis Engine for Time Series Data
Population-calibrated three-layer thresholds for 212 database diagnostic rules complement declarative RCA engines
Drafted by: agent draft (gemini)
operationalizes → Most Health Rules Do Not Know Their Fleet: Population-Calibrated Thresholds, Three-Layer Effective Defaults, and Stratified Epidemiology of 212 Database Diagnostic Rules → Soufflé: On Synthesis of Program Analyzers
Applies Soufflé's compiled-Datalog synthesis pattern to population-calibrated fleet diagnostic rulepacks.
Drafted by: agent draft (gemini)
complements ↔ Machine-Checked Log Compression: Zero-Axiom Lean 4 Verification and C-ABI Differential Testing for Streaming Log Parsers and Temporal RLE ↔ Can Large Language Models Write Good Property-Based Tests?
Combines proptest C-ABI differential property-based testing and cargo-mutants with zero-axiom Lean 4 formal proofs.
Drafted by: agent draft (gemini)
operationalizes → Machine-Checked Log Compression: Zero-Axiom Lean 4 Verification and C-ABI Differential Testing for Streaming Log Parsers and Temporal RLE → QuickCheck
Uses proptest random generators over C-ABI FFI to differentially test Rust glog parsing against Lean 4.
Drafted by: agent draft (gemini)
operationalizes → Machine-Checked Log Compression: Zero-Axiom Lean 4 Verification and C-ABI Differential Testing for Streaming Log Parsers and Temporal RLE → An Analysis and Survey of the Development of Mutation Testing
Applies cargo-mutants across 121 workspace mutants (killing 72/72 semantic data-plane mutants) alongside Lean 4 proofs.
Drafted by: agent draft (gemini)
contrasts with ↔ What The Log: Compile-Time AST Template Mining and Two-Tier Statistical Temporal Compression for Agent-Queryable Parquet Log Lakehouses ↔ LILAC: Log Parsing using LLMs with Adaptive Parsing Cache
Replaces LILAC's runtime LLM-populated template cache with compile-time source-AST template extraction and deterministic fallback shapes.
Drafted by: agent draft (gemini)
extends → What The Log: Compile-Time AST Template Mining and Two-Tier Statistical Temporal Compression for Agent-Queryable Parquet Log Lakehouses → The dataflow model
Extends Dataflow session windowing with Two-Tier Statistical Temporal RLE (AutoEwma log-space IAT threshold + 60s span cap + Welford/quantile/outlier reservoir).
Drafted by: agent draft (gemini)
operationalizes → What The Log: Compile-Time AST Template Mining and Two-Tier Statistical Temporal Compression for Agent-Queryable Parquet Log Lakehouses → Guidelines for assessing the accuracy of log message template identification techniques
Implements Khan et al.'s canonical wildcard post-processing and token-normalization rules to prevent template over-splitting.
Drafted by: agent draft (gemini)
Full vault context
This page lists the 23 drawn papers and 40 semantic relation notes touching Distributed Systems. For the site-wide four evidence chains, 3 explicit refutations, retracted claim, and all 23 authored records, return to The Research Network overview →
Take the data
The graph you are looking at is a static JSON extract of the vault, served from this origin with no runtime requests. Every count on this page is reproducible from these files and scripts/build-research-graph.mjs.