The 3-GeV RCS (rapid cycling synchrotron) in the J-PARC (Japan Accelerator Research Complex) project accelerates a proton beam of 0.333 mA up to 3 GeV. At 25-Hz repetition rate, the RCS generates a high-power beam of 1 MW. In such an accelerated proton beam, neutrons as well as gamma rays are generated. The cumulative energy dose will be of the order of 100 MGy over 30 years of operation. So as to minimize the radiation exposure during maintenance, it is necessary to construct a vacuum system with reliable components which have a long life in such a high level of radiation. In addition, it is necessary to keep the operating pressure of the beam in ultra high vacuum to suppress pressure instability. Thus we should think not only of the outgassing mainly due to ion-induced desorption but also of the pumping efficiency. Based on the above, the vacuum system was designed as follows: The ring is divided by isolation valves into 6 sections (3 straight and 3 arc sections), which can be pumped down independently. To avoid any eddy current loss, ceramic ducts are used in the bending and focusing magnets. These ducts are connected to titanium ducts placing the Ti bellows between. Here, we adopt pure Ti as a material for the ducts and bellows because of its small residual radioactivity. The ring is evacuated with 20 ion pumps (0.7 m 3/s) and 24 turbomolecular pumps (TMPs) (1.3 m 3/s), which are attached to the Ti ducts. The TMPs are used not only for rough pumping but also for evacuation during the beam operation. Especially a collimator system for localizing beam losses in a restricted area is evacuated with the TMPs, because the outgassing from this region will probably be the greatest. In addition, the straight sections for beam injection and extraction are pumped down mainly by the TMPs. Each arc section is pumped by 4 ion pumps and 2 TMPs. To realize the above system, we developed components such as large aperture ceramic ducts and TMPs with high radioactive resistance, as well as several kinds of heat treatment to reduce the outgassing. Finally, we realized the UHV without baking in the RCS and the beam operation has been successful to date.