{"id":15699249,"url":"https://github.com/dunnock/ipc-bench","last_synced_at":"2025-05-09T02:27:15.798Z","repository":{"id":99985224,"uuid":"230260983","full_name":"dunnock/ipc-bench","owner":"dunnock","description":"Rust ipc-channel throughput benchmark between 2 processes","archived":false,"fork":false,"pushed_at":"2019-12-27T21:43:21.000Z","size":53,"stargazers_count":9,"open_issues_count":0,"forks_count":1,"subscribers_count":1,"default_branch":"master","last_synced_at":"2025-05-03T11:04:07.112Z","etag":null,"topics":["bincode","ipc","ipc-channel","performance"],"latest_commit_sha":null,"homepage":null,"language":"Rust","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/dunnock.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}},"created_at":"2019-12-26T12:32:06.000Z","updated_at":"2024-01-06T08:36:39.000Z","dependencies_parsed_at":"2023-05-11T10:45:29.809Z","dependency_job_id":null,"html_url":"https://github.com/dunnock/ipc-bench","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dunnock%2Fipc-bench","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dunnock%2Fipc-bench/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dunnock%2Fipc-bench/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dunnock%2Fipc-bench/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/dunnock","download_url":"https://codeload.github.com/dunnock/ipc-bench/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":253177148,"owners_count":21866274,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","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":["bincode","ipc","ipc-channel","performance"],"created_at":"2024-10-03T19:39:15.988Z","updated_at":"2025-05-09T02:27:15.785Z","avatar_url":"https://github.com/dunnock.png","language":"Rust","funding_links":[],"categories":[],"sub_categories":[],"readme":"Test send and receive throughput for Mozilla's Servo [ipc-channel](https://github.com/servo/ipc-channel) between 2 processes\n\n## start\n```shell\ncargo bench\n```\n\n## Results\n\n## UPDATE!!\n\nIt appears I was missing to tell serde how to work with bytes, following fix makes performance sending with bincode serde comparable to sending raw bytes array. Thanks [Dr-Ermann](https://github.com/Dr-Emann/ipc-bench) for pointing out:\n```rust\n#[derive(Serialize, Deserialize, Clone)]\nstruct Message {\n pub topic: u32,\n #[serde(with = \"serde_bytes\")]\n pub data: Vec\u003cu8\u003e\n}\n```\n\nIpcSender\u003cMessage\u003e / IpcReceiver\u003cMessage\u003e\n\n```log\nsends_custom/1024 time: [1.7104 us 1.7234 us 1.7407 us] \n thrpt: [561.02 MiB/s 566.63 MiB/s 570.95 MiB/s]\n\nsends_custom/10240 time: [4.4657 us 4.5620 us 4.6565 us] \n thrpt: [2.0480 GiB/s 2.0905 GiB/s 2.1355 GiB/s]\n\nsends_custom/51200 time: [43.689 us 44.632 us 45.778 us] \n thrpt: [1.0416 GiB/s 1.0684 GiB/s 1.0914 GiB/s]\n\nsends_custom/102400 time: [69.929 us 70.925 us 72.257 us] \n thrpt: [1.3198 GiB/s 1.3446 GiB/s 1.3638 GiB/s]\n```\n\n```log\nreceives_custom/1024 time: [1.7672 us 1.8164 us 1.8696 us] \n thrpt: [522.35 MiB/s 537.63 MiB/s 552.60 MiB/s]\n\nreceives_custom/10240 time: [4.3900 us 4.4939 us 4.6213 us] \n thrpt: [2.0636 GiB/s 2.1221 GiB/s 2.1724 GiB/s]\n\nreceives_custom/51200 time: [42.596 us 42.876 us 43.201 us] \n thrpt: [1.1038 GiB/s 1.1121 GiB/s 1.1194 GiB/s]\n\nreceives_custom/102400 time: [70.224 us 70.906 us 71.863 us] \n thrpt: [1.3271 GiB/s 1.3450 GiB/s 1.3580 GiB/s]\n```\n\nTest results with IpcBytesSender/IpcBytesReceiver\n\n```log\nsends_custom_bytes/1024 time: [1.5317 us 1.5523 us 1.5793 us] \n thrpt: [618.36 MiB/s 629.09 MiB/s 637.55 MiB/s]\n\nsends_custom_bytes/10240 \n time: [3.6246 us 3.6457 us 3.6703 us]\n thrpt: [2.5983 GiB/s 2.6159 GiB/s 2.6311 GiB/s]\n\nsends_custom_bytes/51200 \n time: [40.257 us 40.851 us 41.614 us]\n thrpt: [1.1459 GiB/s 1.1673 GiB/s 1.1845 GiB/s]\n\nsends_custom_bytes/102400 \n time: [60.408 us 61.474 us 62.816 us]\n thrpt: [1.5182 GiB/s 1.5514 GiB/s 1.5787 GiB/s]\n```\n\n```log\nreceives_custom_bytes/1024 \n time: [1.4920 us 1.5064 us 1.5213 us]\n thrpt: [641.93 MiB/s 648.26 MiB/s 654.51 MiB/s]\n\nreceives_custom_bytes/10240 \n time: [3.6613 us 3.7150 us 3.7777 us]\n thrpt: [2.5245 GiB/s 2.5671 GiB/s 2.6047 GiB/s]\n\nreceives_custom_bytes/51200 \n time: [41.050 us 42.098 us 43.319 us]\n thrpt: [1.1008 GiB/s 1.1327 GiB/s 1.1616 GiB/s]\n\nreceives_custom_bytes/102400 \n time: [57.150 us 57.577 us 58.095 us]\n thrpt: [1.6416 GiB/s 1.6563 GiB/s 1.6687 GiB/s]\n ```\n\n## Summary\n\nBenches compare sending of byte buffer of a given size either within a `Serialize` derived structure vs plain buffer.\n\nIt appears that sending and receiving bytes has about 5-10 times higher throughput than sending serializable object. Surprisingly sending 50KiB is more than 2 times slower than sending 10KiB and 100KiB - should be very platform specific. Sending/receiving bytes buf throughput has high variance. Sending/receiving serializable object has low variance.\n\nTests performed on MacBook Pro '13 (which sucks at perf tests), tests named by pattern `type/message size`. \n\n### Send Message structure\n\nStructure of a form, where size of Message::data variable between tests:\n\n```rust\n#[derive(Serialize, Deserialize)]\npub struct Message {\n\t//pub topic: String,\n\tpub topic: u32,\n\tpub data: Vec\u003cu8\u003e\n}\n```\n\nSend `Message.clone()` via IpcSender (Serialized via bincode)\n\n```log\nsends/1024 time: [7.8511 us 7.9644 us 8.0932 us] \n thrpt: [120.66 MiB/s 122.62 MiB/s 124.39 MiB/s]\n\nsends/10240 time: [61.313 us 61.641 us 61.968 us] \n thrpt: [157.59 MiB/s 158.43 MiB/s 159.28 MiB/s]\n\nsends/51200 time: [310.77 us 314.08 us 318.15 us] \n thrpt: [153.47 MiB/s 155.46 MiB/s 157.12 MiB/s]\n\nsends/102400 time: [600.69 us 605.17 us 610.21 us] \n thrpt: [160.04 MiB/s 161.37 MiB/s 162.57 MiB/s]\n```\n\nReceive `Message` via `IpcReceiver` (Deserialized via bincode)\n\n```log\nreceives/1024 time: [7.6571 us 7.6989 us 7.7505 us] \n thrpt: [126.00 MiB/s 126.84 MiB/s 127.54 MiB/s]\n\nreceives/10240 time: [60.936 us 61.342 us 61.843 us] \n thrpt: [157.91 MiB/s 159.20 MiB/s 160.26 MiB/s]\n\nreceives/51200 time: [311.62 us 316.45 us 322.98 us] \n thrpt: [151.18 MiB/s 154.30 MiB/s 156.69 MiB/s]\n\nreceives/102400 time: [661.13 us 668.65 us 676.87 us] \n thrpt: [144.28 MiB/s 146.05 MiB/s 147.71 MiB/s]\n```\n\n\n\n## IPC bytes channel\n\n```rust\ntype Message = Vec\u003cu8\u003e\n```\n\nSending bytes buf `Vec\u003cu8\u003e.as_slice()`\n\n```log\nbytes_sends/1024 time: [1.4149 us 1.4572 us 1.5078 us] \n thrpt: [647.69 MiB/s 670.16 MiB/s 690.22 MiB/s]\n\nbytes_sends/10240 time: [4.1461 us 4.2852 us 4.4799 us] \n thrpt: [2.1288 GiB/s 2.2255 GiB/s 2.3002 GiB/s]\n\nbytes_sends/51200 time: [57.111 us 58.914 us 60.991 us] \n thrpt: [800.58 MiB/s 828.80 MiB/s 854.97 MiB/s]\n\nbytes_sends/102400 time: [59.126 us 60.230 us 61.648 us] \n thrpt: [1.5470 GiB/s 1.5834 GiB/s 1.6129 GiB/s]\n```\n\nSending cloned bytes buf `Vec\u003cu8\u003e.clone().as_slice()`\n\n```log\nbytes_sends_cloned/1024 time: [1.6410 us 1.6559 us 1.6758 us] \n thrpt: [582.73 MiB/s 589.75 MiB/s 595.09 MiB/s]\n\nbytes_sends_cloned/10240 \n time: [4.0405 us 4.0725 us 4.1074 us]\n thrpt: [2.3219 GiB/s 2.3417 GiB/s 2.3603 GiB/s]\n\nbytes_sends_cloned/51200 \n time: [43.689 us 44.761 us 45.886 us]\n thrpt: [1.0392 GiB/s 1.0653 GiB/s 1.0914 GiB/s]\n\nbytes_sends_cloned/102400 \n time: [70.887 us 72.588 us 74.637 us]\n thrpt: [1.2778 GiB/s 1.3138 GiB/s 1.3454 GiB/s]\n```\n\nReceiving bytes buf as `Vec\u003cu8\u003e`\n\n```log\nbytes_receives/1024 time: [1.3999 us 1.4303 us 1.4700 us] \n thrpt: [664.33 MiB/s 682.75 MiB/s 697.62 MiB/s]\n\nbytes_receives/10240 time: [4.1569 us 4.2248 us 4.3060 us] \n thrpt: [2.2147 GiB/s 2.2573 GiB/s 2.2942 GiB/s]\n\nbytes_receives/51200 time: [38.904 us 42.204 us 45.431 us] \n thrpt: [1.0496 GiB/s 1.1298 GiB/s 1.2257 GiB/s]\n\nbytes_receives/102400 time: [60.481 us 62.754 us 65.247 us] \n thrpt: [1.4616 GiB/s 1.5197 GiB/s 1.5768 GiB/s]\n```\n\n\n## Bincode\n\nIt does seem that main reason of slow send is Serialization/Deserialization, let's see bincode for \n```rust\n#[derive(Serialize, Deserialize)]\npub struct Message {\n\t//pub topic: String,\n\tpub topic: u32,\n\tpub data: Vec\u003cu8\u003e\n}\n```\n\nEncode\n\n```log\nbincode_encode/1024 time: [8.8087 us 9.0109 us 9.2546 us] \n thrpt: [105.52 MiB/s 108.38 MiB/s 110.86 MiB/s]\n\nbincode_encode/10240 time: [87.656 us 89.375 us 91.548 us] \n thrpt: [106.67 MiB/s 109.27 MiB/s 111.41 MiB/s]\n\nbincode_encode/51200 time: [433.69 us 439.43 us 447.46 us] \n thrpt: [109.12 MiB/s 111.12 MiB/s 112.59 MiB/s]\n\nbincode_encode/102400 time: [865.30 us 879.94 us 899.56 us] \n thrpt: [108.56 MiB/s 110.98 MiB/s 112.86 MiB/s]\n```\n\nDecode\n\n```log\nbincode_decode/1024 time: [2.9798 us 3.1397 us 3.3105 us] \n thrpt: [294.99 MiB/s 311.03 MiB/s 327.73 MiB/s]\n\nbincode_decode/10240 time: [29.083 us 31.390 us 34.096 us] \n thrpt: [286.42 MiB/s 311.10 MiB/s 335.78 MiB/s]\n\nbincode_decode/51200 time: [133.91 us 147.23 us 164.21 us] \n thrpt: [297.34 MiB/s 331.64 MiB/s 364.64 MiB/s]\n\nbincode_decode/102400 time: [333.02 us 359.75 us 393.40 us] \n thrpt: [248.24 MiB/s 271.45 MiB/s 293.24 MiB/s]\n```\n\n\n## Further investigation\n\n[ipc_channel_custom.rs](https://github.com/dunnock/ipc-bench/blob/master/benches/ipc_channel_custom.rs) takes it even further, using the same underlying data layer (just a `Vec\u003cu8\u003e`) with \nchannels instantiated either `ipc::channel` or `ipc::bytes_channel`, when sending bytes it uses custom tailored serialization.\n\nAs it can be seen custom tailored Serialization/Deserialization implementation working up to 10x times faster than bincode's.\n\n### bincode's serialization with simple `struct(Vec\u003cu8\u003e)` data:\n\n```\nsends_custom/1024 time: [7.1496 us 7.1883 us 7.2374 us] \n thrpt: [134.93 MiB/s 135.85 MiB/s 136.59 MiB/s]\n\nsends_custom/10240 time: [56.718 us 56.893 us 57.076 us] \n thrpt: [171.10 MiB/s 171.65 MiB/s 172.18 MiB/s]\n\nsends_custom/51200 time: [295.99 us 298.58 us 301.73 us] \n thrpt: [161.83 MiB/s 163.53 MiB/s 164.97 MiB/s]\n\nsends_custom/102400 time: [581.99 us 583.39 us 585.09 us] \n thrpt: [166.91 MiB/s 167.39 MiB/s 167.80 MiB/s]\n```\n\n```\nreceives_custom/1024 time: [7.1560 us 7.2338 us 7.3347 us] \n thrpt: [133.14 MiB/s 135.00 MiB/s 136.47 MiB/s]\n\nreceives_custom/10240 time: [56.625 us 56.965 us 57.326 us] \n thrpt: [170.35 MiB/s 171.43 MiB/s 172.46 MiB/s]\n\nreceives_custom/51200 time: [296.42 us 296.94 us 297.62 us] \n thrpt: [164.06 MiB/s 164.44 MiB/s 164.73 MiB/s]\n\nreceives_custom/102400 time: [588.19 us 590.89 us 593.70 us] \n thrpt: [164.49 MiB/s 165.27 MiB/s 166.03 MiB/s]\n```\n\n### custom tailored serialization based on same `Vec\u003cu8\u003e` data layer:\n\n```\nsends_custom_bytes/1024 time: [1.5784 us 1.6231 us 1.6922 us] \n thrpt: [577.10 MiB/s 601.66 MiB/s 618.70 MiB/s]\nsends_custom_bytes/10240 \n time: [3.6300 us 3.6490 us 3.6735 us]\n thrpt: [2.5961 GiB/s 2.6135 GiB/s 2.6272 GiB/s]\n\nsends_custom_bytes/51200 \n time: [29.473 us 29.964 us 30.615 us]\n thrpt: [1.5575 GiB/s 1.5914 GiB/s 1.6179 GiB/s]\nsends_custom_bytes/102400 \n time: [51.452 us 52.001 us 52.597 us]\n thrpt: [1.8132 GiB/s 1.8339 GiB/s 1.8535 GiB/s]\n```\n\n```\nreceives_custom_bytes/1024 \n time: [1.5848 us 1.6060 us 1.6289 us]\n thrpt: [599.51 MiB/s 608.05 MiB/s 616.20 MiB/s]\n\nreceives_custom_bytes/10240 \n time: [3.6046 us 3.6162 us 3.6305 us]\n thrpt: [2.6269 GiB/s 2.6372 GiB/s 2.6457 GiB/s]\n\nreceives_custom_bytes/51200 \n time: [30.061 us 31.198 us 32.522 us]\n thrpt: [1.4662 GiB/s 1.5284 GiB/s 1.5862 GiB/s]\n\nreceives_custom_bytes/102400 \n time: [53.772 us 55.318 us 57.163 us]\n thrpt: [1.6683 GiB/s 1.7240 GiB/s 1.7735 GiB/s]\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdunnock%2Fipc-bench","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdunnock%2Fipc-bench","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdunnock%2Fipc-bench/lists"}