This paper presents a mechanism that uses off-line profile information to examine when garbage is best collected. This information is then used to guide the garbage collection frequency in order to reduce the garbage collection time and total execution time.