Photoplethysmography (PPG) is simple, non-invasive and unobtrusive, so it has been commonly used for heart rate (HR) sensing during exercise. PPG-based HR sensor can be applied for almost any part of human body where there is an artery close to the sensor, just by stabilizing it on the skin surface with belt. However, in order to make the HR sensor stable even during vigorous exercise, it requires a high fastening belt pressure, which results in discomfort for the sensor wearer. In this paper, we investigate the relationship between fastening belt pressure and accuracy for PPG-based HR sensors put on two positions such as forearm and wrist. First of all, we conducted a preliminary experiment using 10 subjects to associate fastening belt pressure (from 10hPa to 90hPa) to human comfort or discomfort (loose, moderate, tight, and very tight). Then, we conducted an experiment to measure HR for 10 subjects during exercise, changing the belt pressure and exercise intensity. Experimental results reveal that the forearm position gives higher accuracy for HR sensing than the wrist position, however, exercise severely introduces motion artifact (MA) even for the for earm position. Therefore, if we want to achieve higher accuracy during exercise with moderate fastening belt pressure, a technique to cancel MA is required even for forearm-type PPG-based HR sensors.