{"id":28386013,"url":"https://github.com/unum-cloud/udisk","last_synced_at":"2026-02-26T07:48:03.788Z","repository":{"id":128710907,"uuid":"594748334","full_name":"unum-cloud/udisk","owner":"unum-cloud","description":"The fastest ACID-transactional persisted Key-Value store designed as modified LSM-Tree for NVMe block-devices with GPU-acceleration and SPDK to bypass the Linux kernel","archived":false,"fork":false,"pushed_at":"2023-03-31T15:46:46.000Z","size":53,"stargazers_count":74,"open_issues_count":0,"forks_count":1,"subscribers_count":6,"default_branch":"main","last_synced_at":"2025-06-26T12:51:49.168Z","etag":null,"topics":["cuda","database","io-uring","iouring","key-value","key-value-store","linux","linux-kernel","lsm-tree","spdk"],"latest_commit_sha":null,"homepage":"https://unum.cloud/ustore","language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/unum-cloud.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2023-01-29T14:26:14.000Z","updated_at":"2025-06-16T10:13:25.000Z","dependencies_parsed_at":"2023-06-04T21:15:07.594Z","dependency_job_id":null,"html_url":"https://github.com/unum-cloud/udisk","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/unum-cloud/udisk","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unum-cloud%2Fudisk","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unum-cloud%2Fudisk/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unum-cloud%2Fudisk/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unum-cloud%2Fudisk/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/unum-cloud","download_url":"https://codeload.github.com/unum-cloud/udisk/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/unum-cloud%2Fudisk/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29851825,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-25T22:37:40.667Z","status":"online","status_checked_at":"2026-02-26T02:00:06.774Z","response_time":89,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cuda","database","io-uring","iouring","key-value","key-value-store","linux","linux-kernel","lsm-tree","spdk"],"created_at":"2025-05-30T12:38:22.427Z","updated_at":"2026-02-26T07:48:03.783Z","avatar_url":"https://github.com/unum-cloud.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# UDisk\n\nThe fastest ACID-transactional persisted Key-Value store designed for NVMe block-devices with GPU-acceleration and SPDK to bypass the Linux kernel.\n\n## Configuration Parameters\n\nMost important parameters:\n\n- `directory`: Where to keep the configs, schemas, data, and logs? The boot drive?\n- `memory_limit`: How much RAM to use? 80% of it?\n- `value_max_size`: How large the values can get?\n\nTuning performance?\n\n- `directories_data`: Want to separate large data and put it on some RAID array?\n- `gpu_devices`: Do you a have a GPU we can borrow for acceleration?\n\n### Versioning\n\n```json\n\"version\": 1\n```\n\n### Disk IO\n\nPath where there database would store metadata:\n\n```json\n\"directory\": \"/etc/udisk/\"\n```\n\nPaths where the database would store data, with tuples containing the path and a limit:\n\n```json\n\"data_directories\": [\n    {\n        \"path\": \"/mnt/md0/ukv/\",\n        \"max_size\": 1e12,\n    },\n    {\n        \"path\": \"/mnt/md1/ukv/\",\n        \"max_size\": 1e12,\n    }\n]\n```\n\nChoose the underlying IO mechanism:\n\n```json\n\"io_mechanism\": \"posix\"\n\"io_mechanism\": \"pulling\"\n\"io_mechanism\": \"polling\"\n```\n\nFor asynchronous IO mechanisms you can set the queue depth:\n\n```json\n\"io_queue_depth\": 4096\n```\n\n### Random Access Memory\n\nLimit the total amount of RAM available to the database:\n\n```json\n\"memory_limit\": 1000000000\n```\n\nLimit the amount of memory to be used for caching.\nGreatly affects the performance of single reads.\nDoesn't affect the performance of batch reads.\n\n```json\n\"cache_limit\": 100000000\n```\n\nLimit the maximum number of entries to be collected in-memory (and in the Write Ahead Log), before flushing them into files.\n\n```json\n\"write_buffer_max_elements\": 100000\n```\n\nLimit the maximum combined size of entries to be collected in-memory (and in the Write Ahead Log), before flushing them into files.\n\n```json\n\"write_buffer_max_bytes\": 100000000\n```\n\n### Hardware Acceleration\n\nThreads to be used for compactions, non-blocking garbage collection, and polling mechanisms.\nFor optimal performance with high-end NVMe SSDs we suggest reserving 2 or 3 threads per SSD.\n\n```json\n\"threads\": 1\n```\n\nDefine, how many threads may be accessing the database concurrently:\n\n```json\n\"concurrency_limit\": 64\n```\n\nOptional array of integer identifiers for the GPUs to be used for background compactions:\n\n```json\n\"gpu_devices\": \"all\"\n\"gpu_devices\": [-1]\n\"gpu_devices\": [0, 1, 2, 3]\n```\n\nThe amount of Video RAM available to UDisk on the GPU:\n\n```json\n\"gpu_memory_limit\": 1000000000\n```\n\n### Structure\n\nDefine, how large the biggest value can be:\n\n```json\n\"value_max_size\": 4096\n```\n\nIn some rare cases, you may have all of your values of the same size:\n\n```json\n\"value_size\": 0\n```\n\nDefine, how much larger every layer of the Log-Structured-Merge Tree can get:\n\n```json\n\"level_enlarge_factor\": 4\n```\n\nHow large the top-level of the Log-Structured-Merge Tree should be?\n\n```json\n\"first_level_max_bytes\": 1000000000\n```\n\n### Transactions\n\nThe maximum number of entries to be updated with a single transaction:\n\n```json\n\"transaction_max_elements\": 100000\n```\n\nThe maximum capacity of a single transaction, limiting the size of all updates in it:\n\n```json\n\"transaction_max_bytes\": 100000000\n```\n\n---\n\n### Full Example\n\n```json\n{\n    \"version\": 1,\n    \"threads\": 1,\n    \"memory_limit\": 100000000000,\n    \"gpu_devices\": [-1],\n    \"gpu_memory_limit\": 1000000000,\n    \"directory\": \"./tmp/\",\n    \"data_directories\": [\n        {\n            \"path\": \"/mnt/disk1/\",\n            \"max_size\": 100000000000\n        },\n        {\n            \"path\": \"/mnt/disk2/\",\n            \"max_size\": 100000000000\n        }\n    ],\n    \"io_mechanism\": \"pulling\",\n    \"io_queue_depth\": 4096,\n    \"transaction_max_elements\": 1000,\n    \"transaction_max_bytes\": 13421773,\n    \"concurrency_limit\": 128,\n    \"cache_limit\": 134217728,\n    \"write_buffer_max_elements\": 100000,\n    \"write_buffer_max_bytes\": 134217728,\n    \"first_level_max_bytes\": 4831838208,\n    \"level_enlarge_factor\": 4,\n    \"value_size\": 0,\n    \"value_max_size\": 4194304\n}\n```\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Funum-cloud%2Fudisk","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Funum-cloud%2Fudisk","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Funum-cloud%2Fudisk/lists"}