Parallel Gaussian Process Regression with Low-Rank Covariance Matrix Approximations