Loading…
Topic
812 posts about Performance. Every summary links to the original.
See the idea
Bare metal gives engineers direct control over GPU, network, and virtualization layers, but it shifts elasticity and lifecycle work onto the team. This trade-off map shows when that bargain pays off.
Bun packs a runtime, package manager, bundler, and test runner into one binary. Here is how those four parts fit together and where the seams still show.
A five-step how-to for deciding when an AMD Instinct GPU should stay dedicated and when SR-IOV virtual functions, VMs, or sidecar isolation can safely share it.
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
Newest first
React Native Skia exposes Google’s cross-platform Skia 2D graphics engine to React Native, giving developers a canvas for custom shapes, visual effects, and dynamic interactions. The article explains when to choose it over react-native-svg or standard React Native Views: SVG is optimized for static SVGs, while Skia supplies primitives such as shaders, path operations, and image filters for animated experiences. It introduces the Cartesian canvas, basic Circle, Rect, and RoundedRect shapes, and custom paths created with SVG notation or imperative commands including addCircle, lineTo, and addPoly. Examples then combine Skia with react-native-reanimated shared values for UI-thread animation and react-native-gesture-handler for panning, before adding Blur and DashPathEffect; the article also notes web support through React Native Web and points to further examples.
Grab needed to localise nearly 30 million road segments across more than 10,000 area boundaries in Southeast Asia to attach hyperlocal attributes like language, driving side, and vehicle access rules. Testing direct geometric inclusion between complex road polylines and boundary polygons proved computationally prohibitive for daily map generation pipelines. To optimize this process, Grab introduced geohashes as rectangular proxies, precomputing geohash coverage across borders and road segments before joining the datasets in parallel. To resolve misclassification errors near borders without sacrificing performance, the team evaluated geohash coverage percentages and refined boundary-touching geohashes with polygon clipping. A laptop benchmark on a subset of data reduced runtime from 38 minutes with geometric intersection to 78 seconds using the geohash approximation.
Ruvy is an open-source toolchain that takes Ruby code as input and produces a WebAssembly module whose exported _start function executes that code. Built on ruby.wasm, it pre-initializes the Ruby virtual machine and includes Ruby files during the build, while avoiding the need to supply a Ruby script path as a WASI argument at runtime. In Wasmtime benchmarks, Ruvy’s instantiation and _start execution took about 44.4–45.2 ms, compared with 55.8–56.9 ms for ruby.wasm plus wasi-vfs, and its Wasm-to-native compilation took about 440–457 ms versus 1.62–1.68 seconds. These results correspond to roughly 20% faster runtime performance and about 70% less compilation time in the reported tests. Ruvy currently requires users to install build dependencies and compile it because precompiled binaries are not shipped.
Grab's legacy lookalike audience generation platform suffered from long creation SLAs of two working days, high costs, and low weekly update frequencies. To resolve these bottlenecks, the engineering team designed an embedding-based platform powered by an in-memory retrieval service and automated update pipelines. The system creates audience representations by averaging constituent passenger embeddings, determining user membership through real-time cosine score thresholds. To eliminate feature store latency while fitting all embeddings into memory, a hash-based compression method cuts passenger embedding storage needs by roughly 90%. Consequently, audience availability dropped to within 15 minutes of campaign creation, audience generation costs fell by 98%, and ad impressions and clicks doubled.
Grab's Segmentation Platform previously stored user-to-segment mappings across individual rows in ScyllaDB, causing write bottlenecks during segment creation and read latencies too high for downstream consumers. To resolve these performance limitations, the team transitioned to storing segments as Roaring Bitmaps saved as single blobs in object storage. This compression strategy splits 32-bit integer user IDs into chunks across array, bitmap, and run containers based on data density, reducing a one-million-member segment to under one megabyte. An accompanying client SDK manages segment retrieval, decoding, update notifications, and least-recently-used in-memory caching. Consequently, consumers such as Grab's communications platform achieved peak throughput of 15,000 queries per second with sub-millisecond p99 read latencies.
The article provides a tutorial for recreating Supabase Launch Week 8’s animated constellation in Next.js with Three.js and React Three Fiber (R3F). Each signup or user-generated ticket becomes a particle shaped into an “8,” with signups loaded from the `lw8_tickets` table and later updates delivered through Supabase Realtime. The scene uses five-sided circle geometry, a white standard material with optional AdditiveBlending, an R3F Canvas, and a camera positioned at `[0, 0, 500]`. Particle movement is calculated with `useFrame`, elapsed time, sine and cosine functions, alternating loop direction, randomized speed and delay, and configurable offsets; a `dat.gui` debug mode exposes parameters such as particle count and shape dimensions.
Embeddings support semantic search, retrieval augmented generation, and clustering in pgvector, but high-dimensional vectors increase storage, memory, and distance-computation costs as datasets grow. The post explains how IVF indexes reduce the records examined and compares embedding models using Massive Text Embedding Benchmark results, noting little correlation between dimension count and performance. In a 224,482-vector Wikipedia dataset, all-MiniLM-L6-v2 with 384 dimensions delivered 78% higher queries per second than OpenAI text-embedding-ada-002 with 1536 dimensions at accuracy@10 of 0.99, while using 4GB rather than 7.5GB of RAM. The 384-dimensional gte-small model produced more than a 200% queries-per-second improvement over all-MiniLM-L6-v2 at the same accuracy, though model choice still depends on task performance, sequence length, language support, and operational costs.
Supabase’s July 2023 update lists product releases, community content, and preparations for Launch Week 8, which was scheduled to begin the following week. The shipped changes include a new Android Kotlin client library, Login with Figma as an OAuth provider, Resend support as a custom SMTP provider, pgsodium 3.1.8, and PostgREST 11.1. A performance benchmark examines pgvector 0.4.0’s strengths and limitations, while noting that work on HNSW had started and was not far away. The update also features a Snowflake natural-language query application built with Python, Streamlit, Open AI, and Supabase Vector, plus tutorials, courses, community projects, and hiring announcements.
Sidekick addresses two streaming problems in LLM chatbots: Markdown fragments can render awkwardly before syntax is complete, and tool-driven responses can delay visible output across multiple LLM roundtrips. Its solution combines a buffering Markdown parser with an event emitter that multiplexes the main response and asynchronously resolved content into one stream. A Node.js Transform stream processes Unicode characters through a finite-state machine, buffering ambiguous sequences such as emphasis, list markers, and links until they resolve or are flushed as raw text. The parser also recognizes card links, triggering asynchronous card-resolution tasks while Server-Sent Events carry named streams to the UI, which demultiplexes them into the main response and placeholders. The design uses Markdown as a human-readable transport between the LLM, backend, and frontend, allowing content to appear as it is produced.
Llama-2-70B is evaluated against GPT-3.5-turbo on serving cost and latency, with the comparison deliberately excluding GPT-4 and noting fine-tuning as a separate reason to choose Llama. The evaluation serves Llama on two 80-GB A100 GPUs in 16-bit precision, then uses transformer FLOP, memory-bandwidth, and KV-cache analysis to explain the difference between prompt processing and token generation. The measurements indicate that Llama-2 has worse completion-token pricing than GPT-3.5 on two A100s, while its prompt-token pricing is more than three times cheaper; eight A100s might improve pricing but introduce unacceptable latency. The resulting guidance favors GPT-3.5 for most workloads and open-source Llama models for prompt-heavy tasks such as classification or reranking, with batch processing and workloads lacking prompt tokens identified as additional cases where Llama may make sense.
Adding a WebAssembly instruction to Winch, Wasmtime’s baseline non-optimizing compiler, requires coordinated changes across code generation, register management, ISA-specific emission, and tests. The contribution described adds visitor support, delegates through CodeGenContext and MacroAssembler, emits x86_64 instructions through Cranelift-backed assembler methods, and registers the instruction for differential fuzzing. CodeGenContext keeps the value stack, register allocator, and function frame consistent, while spilling can move values to linear-memory offsets; MacroAssembler handles ISA differences, register constraints, and CPU feature fallbacks such as lzcnt versus bsr-based clz. Filetests inspect generated machine code without executing it, while differential fuzzing compares Winch results with another WebAssembly engine across randomly generated modules and inputs. The workflow also includes running fuzzing, incorporating draft-PR feedback, rebasing, and submitting the public Wasmtime pull request.
The post benchmarks pgvector 0.4.0 for approximate nearest-neighbor search and assesses whether PostgreSQL can support production-scale vector workloads. Using the ANN Benchmarks methodology, pre-warmed RAM, the dbpedia-entities-openai-1M dataset of 1 million 1,536-dimensional OpenAI embeddings, and Supabase ARM instances, it varies probes, lists, concurrency, and compute size. On an 8-core, 32 GB instance, pgvector reached 0.91 accuracy@10 at 380 QPS with 10 probes and 0.98 accuracy at 140 QPS with 40 probes; a 64-core, 256 GB server reached about 1,800 QPS at 0.91 accuracy. The results indicate predictable scaling, while higher lists can improve query speed at the cost of index-build time, and the roadmap identifies HNSW, product quantization, and parallel index builds as planned improvements.
PostgREST v11.1 is available on the Supabase platform with configuration and querying improvements, plus a small set of documented breaking changes from v10. It lets JWT-authenticated impersonated roles such as anon, authenticated, and service_role receive ALTER ROLE settings, including statement timeouts, pg_plan_filter statement cost limits, and configurable transaction isolation levels. The release also supports bulk JSON inserts that preserve column defaults when defaultToNull is false, and adds any/all modifiers for several filters, including eq, like, ilike, and comparison filters. For HTTP clients outside Supabase libraries, Range is now ignored on non-GET requests and RPC requests no longer use the deprecated Prefer: params=multiple-objects header; existing Supabase projects can upgrade by pausing and unpausing.
Grab's 69.3 GiB multi-module Go monorepo caused commands like go get to take over 18 minutes as Git repeatedly traversed commit history, downloaded large worktrees, and overloaded their GitLab VCS infrastructure. To bypass direct VCS queries without losing automatic updates for external repositories, the team deployed the Athens Go module proxy configured in fallback network mode. They used the GOVCS environment variable to disable Git access specifically for the monorepo path, forcing Athens to fall back to its internal object storage when resolving monorepo modules. A dedicated CI pipeline pre-populates and refreshes the Athens cache whenever new monorepo modules are released. This setup reduced monorepo go get execution times to approximately 12 seconds and allowed a 70% scale-down of the Athens proxy cluster.
Retrieval plugins let ChatGPT access relevant content from private datasets by chunking documents, generating embeddings with OpenAI’s text-embedding-ada-002, and querying a vector database. The article describes Postgres and Supabase datastore implementations using pgvector, with a shared documents schema and a match_page_sections function that filters by source, author, document, and date before ranking vectors with inner product distance. It then gives a local example: downloading the Postgres 15 documentation as a PDF, uploading it through the plugin’s /upsert-file endpoint, and configuring environment variables for Supabase or Postgres. The example produced 7,904 document records, and an optional IVFFlat index can trade exact nearest-neighbor search for approximate search; after registering the local plugin in ChatGPT, users can ask questions answered from the documentation.
Grab's real-time stream processing platform encountered severe consumer lag and CPU throttling when running Go-based Kafka consumer pipelines on Kubernetes. The issue originated when the Vertical Pod Autoscaler (VPA) scaled pod CPU allocations down to floating-point values such as 1.94 cores. Because AUTO-GOMAXPROCS rounds non-integer CPU limits down to integers, Go runtime thread allocation dropped to 1 core, significantly throttling pipeline throughput despite available pod capacity. Setting a minimum floor of 2 cores instantly restored CPU utilization to 95% and cleared the message backlog. To prevent similar throttling, the team utilized integer CPU scaling recommendations available in VPA v0.13 on Kubernetes 1.25 and above.
Upgrading to Xcode 13.1 introduced severe CI test instability and high CPU utilisation for Grab's iOS development team. To address this, the team integrated observability tools across their UITest pipeline to pinpoint performance bottlenecks and test flakiness. Interventions included isolating spotlight.app to curb CPU spikes, replacing Safari with a mock browser for deep link tests, and booting simulators with pre-granted permissions. The team also built custom network tracking tools to enforce resource mocking and transitioned tests away from arbitrary sleep commands to explicit wait wrappers. These combined changes cut CI runtimes, decreased CPU utilisation by over 50%, and stabilized automated test executions.
Talaria, an open-source distributed time-series database developed at Grab, previously allowed users to run custom data transformation scripts during ingestion using a Lua virtual machine. Launching and executing Lua scripts caused significant performance overhead when processing large volumes of events. To resolve this bottleneck, the team replaced the Lua VM with Go plugins compiled as Linux shared libraries (.so files). Benchmarks revealed that calling Go plugins achieves performance on par with native Go functions, executing roughly 2.3 times faster and consuming 2.3 times less memory than cached Lua VMs. Both execution methods conform to a unified Handler interface to load and run custom transformations.
PostgreSQL pluggable storage would let different tables in one database use storage engines optimized for transactional, analytical, or archival workloads, while allowing developers to build engines in an extensions-like model. PostgreSQL 12 added basic support through refactoring and expanding table access method APIs, initially to enable ZHeap, whose undo log showed promising results and addressed long-term table bloat, but work on ZHeap and TAM appears stalled. The current API has important gaps, including limited index flexibility, all-or-nothing index updates, and row-version identifiers tied to block and offset numbers; available options such as Citus Columnar Storage remain constrained by these limits. OrioleDB's upstreaming has reduced its fork's patchset from about 5,000 lines in PostgreSQL 14 to about 2,000 for PostgreSQL 16, raising the possibility that required changes could land in PostgreSQL 17.
pg_graphql v1.2 introduces several API features for projects using PostgreSQL 15.1.0.63 or higher, including reflection of views, materialized views, and foreign tables. Reflected views require a primary-key constraint supplied through a comment directive, while simple views also support insert and update mutation events. Filtering gains null and non-null checks plus like, ilike, and startsWith options; startsWith can use simple B-Tree indexes, while citext and PGroonga address scalability concerns for text filtering. The release adds functional @skip and @include directives and lets users define GraphQL descriptions for tables, views, and columns through comments. Planned work includes user-defined functions, nested inserts, and an alternative computed-relationships approach based on SQL functions returning SET OF.
The best engineering reads, once a week.
