Tool for launching a Linux process from a snapshot
Switch branches/tags
Nothing to show
Clone or download
Mark Seaborn
Mark Seaborn Cleanup: Remove commented-out assertion
This arg is now checked in CanHandleSyscall().
Latest commit fdfc94d Sep 22, 2015
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
tests
LICENSE
README.md
TODO.md
elf_loader.c
elf_loader_linker_script.x
linux_syscall_support.h
make.sh
ptracer.cc
restore.cc

README.md

snappy-start: Tool for process startup snapshots

snappy-start is a tool which takes a snapshot of a Linux program's process state after it has started up. It allows multiple instances of the program to be quickly launched from the snapshot.

This has two potential benefits:

  • Faster startup, if the program does a non-trivial amount of computation during startup.

  • Saving memory, because memory pages that the program writes to during startup will be shared between the instances.

Usage

First, build the tool by running make.sh (which also runs some tests).

To create a snapshot:

./out/ptracer ./out/elf_loader PROG ARGS...

To run the snapshot:

./out/restore

The program will be snapshotted when it first calls an unhandled syscall, such as getpid().

Currently, the snapshot data is written to hard-coded files out_info (containing register state and a list of memory mappings) and out_pages (data to restore using mmap()).

Example:

$ ./out/ptracer ./out/elf_loader /usr/bin/python tests/python_example.py
$ ./out/restore
Hello world, from restored Python process

Credits

The idea for this tool comes from Kenton Varda, who proposed using a "record/replay" approach, using ptrace() to monitor syscalls so that they can later be replayed.