splicy is an experimental splice(2) -based in-memory HTTP caching proxy server.

This is mostly a toy/proof-of-concept project, and should be viewed as such; there is no actual backend, HTTP proxy headers or cache management. Likewise, the performance is most likely horrendous. Use at your own risk.

splicy is an application of some of the concepts exposed by Linus Torvalds when he gave his high-level description of the Linux-specific splice(2) and tee(2) system calls, which can be found here: http://kerneltrap.org/node/6505

Instead of using regular send(2) / recv(2) system calls and userspace virtual memory (such as the one obtained using malloc(2) or mmap(2)) to proxy and store the cached HTTP responses from its backends, splicy uses splice(2) / tee(2) and the kernelspace memory provided by pipe(2). The idea is to try and achieve true zero-copy operation, as the kernel is able to just move/duplicate data without actually copying it.

splicy is © 2012 Nicolas Noirbent, and is available under the AGPL3+ license.

splicy uses the autotools for its build system.

If you checked out code from the git repository, you will need autoconf and automake to generate the configure script and Makefiles.

To generate them, simply run:

$ autoreconf -fvi

You will need Cython to build the Python binding to splice(2) and tee(2).

You need to be able to build Python extensions to build splicy. On most distributions this means installing the python-dev or python-devel package.

splicy builds like your typical autotools-based project:

$ ./configure && make && make install

You will need Linux >= 2.6.17, glibc >= 2.5, Python >= 2.6 and cyhttp11 to run splicy.