As a developer and operator of a distributed file system, and specifically one that might have caused Rachel some pain, I kind of agree with her. If you can run with local code and data, do so. Even if you think you can't, think again. Only run from data on the servers if you have no choice, e.g. if the dataset is too large to copy or you need simultaneous access (better be read-only by that point) from multiple clients. Not only will you be happier, but so will the people who have to operate that distributed filesystem in the face of many users all abusing the hell out of every operation in the POSIX/VFS API.