===========================
Overview
===========================
Anticipating Invariant (AI) is a novel invariant that can
anticipate the concurrency bugs (CBugs) right before their turning points.
It enables the users to forestall the failures caused by CBugs
with a simple thread stalling technique, rather than resort to roll-back.
In this package, we present:
1) the source code of our LLVM-based AI implementation (`ai/*`);
2) several demos for demonstrating AI's ability of tolerating CBugs (`demo/*`);
3) applications from different categories (desktop, server, hpc) for evaluating AI's overhead (`performance/*`);
and 4) an example of the APIs' usage (`api/*`).
Documentations, screencasts and some auxiliary scripts are also provided.
===========================
Documentation
===========================
The documentations of AI are distributed in:
1) the files under the `doc/` directory;
2) the `README` file of each subdirectory;
and 3) the annotations in the `Makefile` file.
A recommended oder of reading these documentations is given below:
+-----------------------------+
| docs/BUILD | # How to build AI.
+-----------------------------+
|
v
+-----------------------------+
| docs/USAGE | # How to use AI.
+-----------------------------+
|
v
+-----------------------------+
| demo/*/README | # The procedure of building and running demos.
+-----------------------------+
|
v
+-----------------------------+
| performance/pbzip2/README | # AI's overhead on desktop application.
+-----------------------------+ # How to relax BSets.
|
v
+-----------------------------+
| performance/apache/README | # AI's overhead on server application.
+-----------------------------+
|
v
+-----------------------------+
| performance/fft/README | # AI's overhead on hpc application.
+-----------------------------+ # Bias instrumentation.
|
v
+-----------------------------+
| api/README | # The usage of AI's APIs.
+-----------------------------+