Reducing planned outages for book hardware maintenance with concurrent book replacement

Abstract

The IBM System z9e introduces the enhanced book availability (EBA) feature to reduce the number of planned system outages. Included as part of the EBA feature is the concurrent book replacement (CBR) function, which allows a single book in a multibook server to be concurrently removed from the system in order for service personnel to perform a repair or to physically upgrade the hardware on the book. This repaired or upgraded book is then concurrently replaced and reintegrated into the server configuration. In this paper, we describe the benefits that the concurrent book replacement function offers a customer during a planned repair or upgrade of the System z9. We also describe a tool, developed to analyze the server in order to determine whether the server is ''prepared'' for the concurrent book replacement operation, and we provide an overview of the actions performed by this tool. The paper also contains a description of the concurrent book replacement operation, with an emphasis on the unique functions developed as part of this design.

Advancing reliability, availability, and serviceability with the IBM z13

“…Failures that require replacement of a drawer or access to the inside of the drawer use concurrent drawer repair (CDR). This replaces concurrent book repair on prior z Systems (with drawer substituted for book) [11]. The workload is temporarily moved to other drawer(s) during the repair.…”

Regression Testing Based on UML Design Models

“…This technique cannot restore lost data and cannot recover from failures that affect unrelocatable pages. In contrast, hardware hypervisor-based systems can perform flexible remapping and migration of OSvisible physical addresses memory pages to new absolute memory locations at runtime [1]. However, this approach requires architectural changes in all levels of the memory system to support page migration and dedicated hardware and firmware to map and manage the memory locations.…”

“…"Memory page retirement" [19] leverages OS virtual memory support to unmap damaged physical pages, but cannot tolerate faults that affect unrelocatable pages [9,11]. Alternatively, the absolute location of physical addresses can be managed by a hypervisor, however this requires significant changes to the system architecture and hardware and firmware changes throughout the design to efficiently support the memory reorganization [1]. Furthermore, because single nodes can be repaired before additional failures occur [16], simple reconstruction mechanisms can provide the same reliability benefits as a heavyweight, but flexible hypervisor architecture.…”

Redundant I/O interconnect

“…Another paper published in this issue focuses on a category of service actions that allow the customer to add or replace processor books to the processor cage of a system-the enhanced book availability (EBA) features that reduce planned outages for book hardware upgrades and maintenance [1]. This approach first requires a solution to the problem presented by a limitation that originates from the classical organization of the self-timed interface (STI) network [2].…”