Different kinds of variations are often used as a measurement of how irregular/rough functions are and they are closely related to various integration theories which can be used to define differential equations driven by such functions. For example, a function is of finite total variation if and only if it induces a (signed) measure, in which case, one can define integrals with respect to that function in the usual sense. However, most stochastic processes in continuous time do not have sample paths of finite total variation and hence different kinds of variations come into play (p-variations and phi-variations). In this talk we will mainly focus on Gaussian processes and describe how to obtain sharp bounds for their variations.