Hypervisor Based Method for Ensuring Consistent and Reliable Access to Shared Resource between Logical Partitions
Publication Date: 2008-Nov-29
The IP.com Prior Art Database
Virtualization compartmentalizes the various software blocks into virtual machines (VM) or logical partitions (LPARs) without compromising the deterministic characteristics of legacy blocks. The VMs are decoupled from the hardware with the isolation of the hypervisor, natively residing atop the silicon. The hypervisor presents the VMs with a suitable set of hardware resources, from number of cores to run on, to I/O available for their usage, to RAM and MMU tables. These hardware resources can be partitioned or virtualized for shared usage across multiple VMs depending on the application scenario. Now, considering a case where such LPARs are sharing a memory region in the RAM. If the LPAR responsible for updating the resource goes offline, the shared resource could be left in an inconsistent and unreliable state. At this point if another LPAR accesses this resource, it may end up performing an undesirable operation based on the data obtained from there. This paper presents a hypervisor based solution to prevent such a problem scenario. The state of the shared resource is linked to the LPAR responsible for updating it i.e. if the writer LPAR goes offline the resource is no longer considered reliable. The hypervisor is able to detect the failure of the writer LPAR and subsequently prevents the dependent LPARs from accessing the shared resource. Access to the shared resource is restored once the writer LPAR recovery is complete. The recovery is also monitored by the hypervisor.