Hasty Briefsbeta

All tags

#performance

528 stories total

Bilingual

Visual Studio Code: Text Buffer Reimplementation (2018)
a year ago
- VS Code 1.21 introduced a new text buffer implementation for better performance and memory usage.
- The previous implementation used an array of lines, which was inefficient for large files.
- A piece table data structure was chosen to reduce memory usage and improve performance.
- The piece table was enhanced with caching for faster line lookups and a balanced binary tree for efficient searches.
- Memory usage was significantly reduced, and file opening times improved with the new implementation.
- Editing performance was stable across different file sizes, though line-based lookups were slower in some cases.
- The team considered but decided against a native C++ implementation due to performance trade-offs.
- Future optimizations include reducing unnecessary calls to getLineContent and improving the Find command.
Gemini 2.5 Flash Preview 05-20
a year ago
- Gemini 2.5 Pro is the most powerful model with maximum response accuracy and state-of-the-art performance, optimized for complex coding, reasoning, and multimodal understanding.
- Gemini 2.5 Flash offers the best price-performance ratio, with well-rounded capabilities and adaptive thinking, ideal for low latency, high volume tasks.
- Gemini 2.0 Flash provides next-gen features, improved capabilities, and a 1M token context window, designed for speed and real-time streaming.
- Gemini 2.0 Flash Preview Image Generation supports conversational image generation and editing, with specific limitations in certain regions.
- Gemini 1.5 Pro is optimized for complex reasoning tasks, capable of processing large datasets, including 2 hours of video or 19 hours of audio.
- Imagen 3 is the highest quality text-to-image model, generating detailed images with better lighting and fewer artifacts.
- Veo 2 is a high-quality text- and image-to-video model, capable of generating detailed videos with artistic nuance.
- Gemini 2.0 Flash Live enables low-latency bidirectional voice and video interactions, processing text, audio, and video inputs.
- Gemini Embedding Experimental achieves state-of-the-art performance in code, multi-lingual, and retrieval tasks.
- Text Embedding models measure the relatedness of text strings, widely used in AI applications, with different dimension sizes and rate limits.
"ZLinq", a Zero-Allocation LINQ Library for .NET
a year ago
- ZLinq is a zero-allocation LINQ library for .NET, leveraging structs and generics to eliminate allocations.
- It includes extensions like LINQ to Span, LINQ to SIMD, and LINQ to Tree (FileSystem, JSON, GameObject, etc.).
- ZLinq supports multiple platforms, including .NET Standard 2.0, Unity, and Godot, and has exceeded 2000 GitHub stars.
- The library achieves 100% coverage of all methods and overloads in .NET 10, ensuring 99% behavior compatibility.
- ZLinq implements optimizations beyond allocation reduction, such as SIMD support, to outperform in most scenarios.
- The library's architecture is based on ValueEnumerable<TEnumerator, T> and IValueEnumerator<T>, reducing method calls and struct size.
- ZLinq provides LINQ to Tree for tree structures, with standard implementations for FileSystem, JSON, Unity's GameObject, and Godot's Node.
- The library avoids commercialization, focusing on maintenance and innovation despite the mental burden on the maintainer.
- ZLinq's development was influenced by feedback and contributions from the community, ensuring compatibility and performance improvements.
JavaScript Ecosystem Performance
a year ago
- Removing redundant dependencies and replacing them with modern alternatives.
- e18e (Ecosystem Performance) initiative connects people and projects improving JS package performance.
- Provides visibility to open-source developers' efforts in cleanup, level-up, and speed-up of dependencies.
- Invitation to join projects and connect with like-minded individuals via the e18e Discord server.
- Sponsorships support community work, rewarding contributors or funding larger projects.
- Opportunity to help by sponsoring through Open Collective.
Fast Allocations in Ruby 3.5
a year ago
- Ruby 3.5 introduces significant speed improvements in object allocation, making it up to six times faster.
- Benchmarks compare positional and keyword parameters with and without YJIT, showing consistent speedups.
- Positional parameters see a 1.8x speedup without YJIT and 2.3x with YJIT, regardless of parameter count.
- Keyword parameters show increasing speedups with more parameters, up to 3x without YJIT and over 6.5x with YJIT.
- The optimization involves inlining the `Class#new` method, eliminating stack frames and parameter copies.
- Inlining improves inline cache hits and reduces overhead by avoiding conversions between Ruby and C calling conventions.
- A downside is a small backward incompatibility where the `Class#new` frame is missing in stack traces.
- Memory usage increases slightly due to more instructions at call sites, but the impact is minimal (0.5% growth in instruction sequence size).
Making the rav1d Video Decoder 1% Faster
a year ago
- Performance improvements in rav1d, a Rust port of the dav1d AV1 decoder, were explored.
- Two key optimizations were identified: avoiding zero-initialization of buffers and optimizing equality comparisons for small numeric structs.
- The first optimization involved using MaybeUninit to avoid unnecessary buffer initialization, improving runtime by 1.2 seconds (1.5%).
- The second optimization replaced field-wise equality comparisons with byte-wise comparisons, improving runtime by 0.5 seconds (0.7%).
- These changes reduced the performance gap between rav1d and dav1d by about 30%, with a remaining difference of 6%.
- The approach used profiling to compare performance between the Rust and C implementations, focusing on specific functions and assembly calls.
- The optimizations were implemented without introducing new unsafe code, maintaining safety while improving performance.
Show HN: Pi Co-pilot – Evaluation of AI apps made easy
a year ago
- Pi Scorer is highly accurate and insanely fast.
- It outperforms Deepseek and GPT 4.1 in accuracy.
- Runs at the size and speed of GPT Mini and Gemini Flash.
- Capable of scoring 20+ custom dimensions in under 100 milliseconds.
TypeScript Native Previews
a year ago
- TypeScript Native Previews are now available via npm and VS Code Marketplace.
- The native TypeScript compiler, named 'tsgo', offers a 10x speed-up and supports JSX and JavaScript+JSDoc type-checking.
- Editor functionality in VS Code is in early stages, with completions enabled but other features pending.
- API progress includes a new IPC-based communication layer and a synchronous Node.js module in Rust.
- Known differences include deprecations like 'node' resolution and limited downlevel emit support.
- Future updates will focus on completing compiler features like '--build' and expanding editor support.
Poireau: A Sampling Allocation Debugger
a year ago
- libpoireau library intercepts a small fraction of malloc/calloc calls to generate a heap footprint overview.
- Plans to implement guard pages similar to Electric Fence for additional security.
- Sampling approach minimizes performance impact, making it suitable for production use.
- Offloads complexity to the kernel or external scripts, reducing invasiveness compared to tools like LeakSanitizer or Valgrind.
- Reports old allocations rather than scanning for references, useful for detecting slow heap growth.
- Targets Linux 4.8+ on 64-bit platforms with 4 KB pages.
- Requires GCC-compatible C11 implementation and LD_PRELOAD for instrumentation.
- Uses Linux perf for tracepoint registration and event generation.
- Provides various signals (SIGHUP, SIGUSR1, SIGUSR2) for interacting with poireau.py to get allocation reports.
- Performance overhead ranges from 5-20% for small/medium allocations to ~70% for large allocations.
- Includes code derived from xoshiro 256+ 1.0 and Systemtap's sys/sdt.h, both in the public domain.
Faster Firewalls with Bpfilter
a year ago
- Bpfilter is a project that improves network traffic filtering performance by using BPF (Berkeley Packet Filter).
- It was proposed in 2018 to enhance iptables performance by translating filtering rules into BPF programs.
- Bpfilter consists of three components: the bpfilter daemon, libbpfilter library, and bfcli command-line tool.
- The project supports iptables and nftables, though nftables support is currently broken and planned for refactoring in 2025.
- Bpfilter translates rules into BPF bytecode, which is then loaded into the kernel for efficient packet filtering.
- Performance benchmarks show bpfilter handles larger rulesets more efficiently than iptables and nftables before performance drops.
- Future plans include better nftables support, integration of user-provided BPF programs, and generic sets.
Short alphanumeric pseudo random identifiers in Postgres
a year ago
- Introduction to generating short, alphanumeric identifiers in Postgres as an alternative to UUIDs.
- Use of public_id for external use and id for internal database references.
- Design properties include fixed size, obfuscation, reversibility, and native Postgres capabilities.
- Limitations include no support for case insensitivity or excluding similar-looking characters.
- PL/PgSQL functions for obfuscation, base62 conversion, and deobfuscation.
- Table design with generated public_id column and constraints for uniqueness and length.
- Performance impact: 3.4x slower inserts with public_id generation.
- Alternatives considered: Base32 Crockford, ULID, and NanoIDs.
Starlite V Linux Tablet Updated
a year ago
- Compact and powerful design with Intel Alder Lake N350 processor and 16GB memory.
- Fanless 0db design with versatile connectivity options including WiFi 5, Bluetooth 5.1, Micro HDMI, USB-C, Micro SD, and headphone jack.
- Secure boot flow, lightweight firmware (0.76s POST time), and endless firmware customization via Advanced Configuration interface.
- 2K resolution display (2160x1440) with 300cd/m² brightness and 178° viewing angle.
- Minuscule 65W Gallium Nitride charger, 30% smaller than traditional chargers.
- Easy firmware updates via LVFS (Linux Vendor Firmware Service).
- Open 1-year warranty allowing hardware modifications without voiding.
- Highly customizable hardware and firmware with pre-installed OS of choice.
Show HN: Zasper – A High Performance IDE for Jupyter Notebooks
a year ago
- Zasper outperforms Jupyter Server in all tested metrics, including CPU and RAM usage.
- Zasper uses up to 5X less CPU and 40X less memory compared to Jupyter Server.
- Zasper maintains performance with 64 kernels at 10 RPS per kernel, while Jupyter Server fails at this load.
- Zasper recovers more gracefully from overload conditions and only fails under extremely high loads.
- Zasper is designed for massive concurrency, offering minimal memory footprint and exceptional speed.
- Benchmarking focused on CPU usage, RAM usage, throughput, latency, and resilience.
- Jupyter Server crashes at lower request throughput due to message queue issues and Python's scalability limits.
- Zasper's Go-based architecture provides better performance and scalability compared to Jupyter's Python-based setup.
- Zasper's efficiency improves under higher loads, making it suitable for large-scale deployments.
- Zasper offers improved responsiveness, lightweight operation, cost efficiency, and better scalability.
Rust Coreutils 0.1.0 Release
a year ago
- Rust Coreutils 0.1.0 released with major performance improvements and SELinux support.
- New SELinux integration in commands like cp, ls, mkdir, and stat, with CI testing now available.
- Performance gains in commands such as cat, ls, wc, and tail, matching or exceeding GNU performance.
- Ubuntu plans to integrate Rust Coreutils into its base system.
- This release includes 843 commits from over 60 contributors, including 40 newcomers.
- Improved GNU test suite compatibility with 84.46% pass rate, up from 82.17%.
- Call to action for sponsorships on GitHub to support project growth.
- Numerous command-specific updates and bug fixes across the utilities.
- Dependency updates and code quality improvements across the project.
- Documentation updates and new contributor acknowledgments.
Rust Coreutils 0.1 Released with Big Performance Gains – Matches or Exceeds GNU
a year ago
- Rust Coreutils 0.1 released with major performance gains.
- Performance of commands like wc, ls, cat, tail, and seq matches or exceeds GNU Coreutils.
- Added SELinux integration for commands such as cp, mkdir, ls, and install.
- Improved GNU Coreutils compatibility, now passing 522 tests from the GNU Test Suite (up from 507).
- Ubuntu 25.10 plans to ship Rust-based Coreutils (uutils) by default.
Btrfs to See More Performance Improvements with Linux 6.16
a year ago
- Btrfs performance improvements in Linux 6.16 include extent buffer conversion to xarray, yielding +50% throughput and -33% runtime for metadata-heavy operations.
- Extent IO tree cleanups and optimizations lead to performance gains by avoiding unnecessary searches.
- More efficient extent unpinning during transaction commits provides an estimated 3-5% runtime improvement.
- User-visible changes: removal of deprecated 'nologreplay' mount option, scrub reporting updates, and device stats message restoration.
- Core updates: subpage mode enhancements, zoned mode improvements for ENOSPC handling, scrub bitmap simplifications, and large folio preparations.
- New support added for compression, buffered write, defrag, hole punching, subpage, and send operations.
- Bug fixes include fsync persistence for files with no hard links and rejection of non-nodesize aligned tree blocks.
- Code cleanups: transaction abort placement, removal of struct bio_vec internals usage, extent map simplifications, and enhanced ASSERT() macro.
Improving performance of original dav1d video decoder
a year ago
- Optimized memory organization in CPU cachelines by reducing structure sizes to 64 bytes or less.
- Manually aligned enums to strict values to fit into 1 byte and optimized space usage.
- Compressed 'int' in structures to 'uint16_t' (2 bytes) to reduce memory waste.
- Used 'pahole' to identify and optimize holes in structures, improving cache efficiency.
- Achieved performance improvements: ~3% for 1080p and ~1% for 4K.
- Reduced the size of 'Dav1dFrameContext' from 5648 bytes to 5384 bytes, saving 4 cachelines.
- Benchmarked using 'hyperfine' on old and new servers, showing consistent performance gains.
- Highlighted the importance of data alignment and structure optimization for 64-bit processors.
- Emphasized the practicality of optimizing existing C/C++ projects over rewriting in new languages like Rust.
Calendars, Contacts and Files in Stalwart
a year ago
- Stalwart v0.12 introduces native support for calendars, contacts, and file storage, eliminating the need for third-party integrations.
- The release includes CalDAV, CardDAV, and WebDAV support, enabling seamless management of events, contacts, and documents.
- Shared resources like group calendars and team-accessible file folders are fully supported, enhancing collaboration.
- Improved spam filtering integrates with users' personal address books, reducing false positives and auto-training the Bayesian classifier.
- Performance optimizations include incremental caching and zero-copy deserialization, reducing load and improving response times.
- Cluster coordination is enhanced with adaptable protocols like Eclipse Zenoh for small deployments and Kafka, Redpanda, NATS, or Redis for larger setups.
- Future updates will include CalDAV Scheduling (RFC 6638), event notification alerts, and JMAP support for Calendars, Contacts, and File Storage.
Yes-rs: A fast, memory-safe rewrite of the classic Unix yes command
a year ago
- yes-rs is a memory-safe, blazingly fast rewrite of the Unix 'yes' command written in Rust.
- Original 'yes' command in C is prone to buffer overflows and lacks modern error handling.
- yes-rs features zero-cost abstractions, fearless concurrency, and no unsafe code blocks.
- Install via 'cargo install yes-rs' or build from source.
- Benchmarks show yes-rs is significantly faster than the GNU 'yes' command.
- Project encourages contributions adhering to Rust's memory safety and performance principles.
- Promotion on platforms like Hacker News is suggested to spread awareness.
LumoSQL
a year ago
- LumoSQL is a modification of SQLite, adding security, privacy, performance, and measurement features.
- It supports pluggable backends, including SQLite Btree, LMDB, and Berkeley Database.
- LumoSQL introduces modern encryption (like Attribute-Based Encryption) and per-row checksums for error detection.
- The project is under the MIT license and supported by the NLNet Foundation.
- LumoSQL runs on x86, ARM-32, RISC-V architectures, and various Linux/BSD OSs.
- It uses the 'Not-forking' tool to track upstream changes without forking SQLite.
- Benchmarking is a key feature, allowing performance comparisons across configurations.
- LumoSQL is maintained using Fossil repository, with contributions encouraged via forums.
- The project aims to demonstrate SQLite improvements that may not fit SQLite's conservative approach.
- Current limitations include incomplete backend support and test coverage.

first prev4next

About|Login

#performance

Visual Studio Code: Text Buffer Reimplementation (2018)

Gemini 2.5 Flash Preview 05-20

"ZLinq", a Zero-Allocation LINQ Library for .NET

JavaScript Ecosystem Performance

Fast Allocations in Ruby 3.5

Making the rav1d Video Decoder 1% Faster

Show HN: Pi Co-pilot – Evaluation of AI apps made easy

TypeScript Native Previews

Poireau: A Sampling Allocation Debugger

Faster Firewalls with Bpfilter

Short alphanumeric pseudo random identifiers in Postgres

Starlite V Linux Tablet Updated

Show HN: Zasper – A High Performance IDE for Jupyter Notebooks

Rust Coreutils 0.1.0 Release

Rust Coreutils 0.1 Released with Big Performance Gains – Matches or Exceeds GNU

Btrfs to See More Performance Improvements with Linux 6.16

Improving performance of original dav1d video decoder

Calendars, Contacts and Files in Stalwart

Yes-rs: A fast, memory-safe rewrite of the classic Unix yes command

LumoSQL