Helix Project Test Bed: Towards the Self-Regenerative Incorruptible Enterprise

Funds provided by this grant allowed us to purchase several large compute-servers, which supported intensive experimentation conducted under the auspices of MURI FA9550-07-1-0532. We conducted extensive mutational robustness studies. Given a population of variant programs, created from an original program by applying random mutations, we measure which variants still pass all available test cases and call them neutral. The fraction of all variants that are neutral is defined as the programs mutational robustness. Even when the mutations are restricted to statements executed by the test cases, mutational robustness is surprisingly high, 36.75 on a corpus of programs taken from 22 production software projects, the Siemens benchmark suite, and a few specially constructed programs. Next, we conducted experiments on the assembly-level automated program repair algorithm using a benchmark test suite of ten open source programs. The ASM repair method successfully repaired all of the ten programs on at least one run, with an overall success rate of 0.54 per trial. We also conducted experiments on a recent extension of the automated program repair framework to elf files. This extension would be useful, for example, when working on embedded devices and either the source code is unavailable or there isnt room to store the complete tool chain required for repairs at the abstract syntax tree level. We used the DURIP-supplied computers to conduct many of the elf level experiments, discovering that on the same benchmark set that we can repair five out of the ten programs at the elf level without any access to the source code.