Fast shared disk switching methods on hot-standby system

摘要

We propose two high-availability methods which support shorter switching time of the shared disks in a hot-standby system. The pseudo-offline method traps a device driver call in each I/O request, and then executes the control routine which determines whether it calls the device driver or not. The control group method classifies the shared disks into a few groups, and controls access to all disks in the same group simultaneously. In existing methods, since they are necessary to issue I/O (Input/Output) for switching shared disks, the switching time is proportional to the number of the shared disks. Then it takes ten and several seconds to switch the shared disks and the whole switching time becomes tens of seconds. In the proposed methods, it is not necessary to issue any I/O requests, the switching time does not depend on the number of the shared disks. We implement the proposed methods in HA Booster, a middleware which boosts system availability. We evaluate Hitachi HA Booster Pack for AIX, and verify that it takes only a few milliseconds constantly to switch the shared disks, and the whole switching time becomes shorter than twenty seconds in our evaluation environment.