Catalyzer: Sub-millisecond Startup for Serverless Computing with Initialization-less Booting

dyweb / papers-notebook

:page_facing_up: :cn: :page_with_curl: 论文阅读笔记（分布式系统、虚拟化、机器学习）Papers Notebook (Distributed System, Virtualization, Machine Learning)

https://github.com/dyweb/papers-notebook/issues?utf8=%E2%9C%93&q=is%3Aissue+is%3Aopen+-label%3ATODO-%E6%9C%AA%E8%AF%BB

Apache License 2.0

2.14k stars 250 forks source link

Catalyzer: Sub-millisecond Startup for Serverless Computing with Initialization-less Booting #237

Open gaocegege opened 3 years ago

gaocegege commented 3 years ago

https://ipads.se.sjtu.edu.cn/_media/publications/duasplos20.pdf

来源：网上冲浪

gaocegege commented 3 years ago

本文基于的一个非常重要的观察：kernel 启动的时间在整个应用启动的生命周期里并不是最长的，更长的是 JVM 等运行时启动的时间。这个在业界所有的轻量虚拟化方案中都是一样的：

gaocegege commented 3 years ago

因此作者先用 gVisor 的 restore/recover 功能 https://gvisor.dev/docs/user_guide/checkpoint_restore/ 来用 restore 替换运行时的 init 过程（修改后的版本叫做 gVisor-restore）。这样改动后确实有提高，但是 restore 的成本还是挺高的。

gVisor-restore 的实现方式还不太看得懂，需要学习一下 gVisor 源码，看论文应该是通过一个 syscall 来劫持了调用，做了一些 save/restore 的操作

We add a new syscall in gVisor to trap at the entry point of serverless functions. We use the term, func-entry point, to indicate the entry point of a serverless function, which is either specified by developers or at the default location: the point right before the wrapped program invoking the handler function. The syscall is invoked by the func-entry point annotation and will block until checkpoint operation begins.

gaocegege commented 3 years ago

尽管用 restore 替换了 JVM 等 runtime init，restore 的成本还是挺高的，所以这篇文章最核心的贡献就是通过一些优化，把 restore 的成本再降低了一下。

Catalyzer 一共引入了三个 boot 类型，分别是 cold boot，warm boot 和 fork boot，他们的 latency 是不一样的。其中 cold boot 就是从前面提到的 func-image 直接 restore 一个 sandbox instance 出来。

warm boot 是 restore 部分 image，同时通过共享内存，从在运行的实例获得一些 state 来避免 restore 全部 image 的成本。hot boot 就是直接通过 Catalyzer 引入的一个新的系统原语 sfork 实现，latency 在文章里的说法是 1ms。

整个实现里，一共涉及四个技术，分别是 Overlay Memory，Seperated State Recovery，on-demand I/O reconnection 和 virtualization sandbox

gaocegege commented 3 years ago

Overlay Memory 因为不熟悉 EPT，不明白实现细节。效果如下：

Catalyzer maintains two layered EPTs for each sandbox. The upper one is called Private-EPT, and the lower one is Base-EPT. Private-EPT is private to each sandbox, while Base-EPT is shared and read-only. During a warm boot, Catalyzer directly maps the Base-EPT for the new sandbox with the share-mapping operation.

m-t-d-32 commented 3 years ago

so can you help me to find where the code of the paper is ? thank you

gaocegege commented 3 years ago

作者是 https://github.com/Ddnirvana ，你要不直接问他

cc @Ddnirvana

m-t-d-32 commented 3 years ago

好的，谢谢

------------------ 原始邮件 ------------------ 发件人: "dyweb/papers-notebook" @.>; 发送时间: 2021年8月11日(星期三) 晚上6:35 @.>; @.**@.>; 主题: Re: [dyweb/papers-notebook] Catalyzer: Sub-millisecond Startup for Serverless Computing with Initialization-less Booting (#237)

作者是 https://github.com/Ddnirvana，你要不直接问他

cc @Ddnirvana

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

gaochong1111 commented 2 years ago

请问有找到这篇论文里的代码实现吗？ @m-t-d-32