|
The introduction of logical partitioning (LPAR) technology to IBM eServer™ pSeries systems has greatly expanded the options for deploying applications and workloads onto server hardware. Logical partitioning is a server design feature that provides more end-user flexibility by making it possible to run multiple, independent operating system images concurrently on a single server. IBM is adding to that LPAR capability with the introduction of dynamic LPAR, in which partition resources can be moved from one partition to another without requiring a reboot of the system or affected partitions. This paper will describe these new capabilities and explain the benefits of their use.
To fully utilize dynamic LPAR capabilities, the following are required:
- A POWER4™ processor-based pSeries system, such as the p690, p670, p630, or follow-on
- The October 2002 (or later) system microcode update
- A Hardware Management Console (HMC) at version R3V1.0 or later
- AIX® 5L™,Version 5.2 or later
Not all partitions on a system must be migrated to AIX 5.2 for dynamic LPAR to work, but only those partitions running AIX 5.2 or later will be able to participate in dynamic LPAR operations. These new versions of system microcode and HMC will continue to support AIX 5.1 at its current level of LPAR functionality.
In an LPAR configuration, individual processor, 256MB memory region, and I/O adapter slot resources are placed under the exclusive control of a given logical partition. One of the main advantages of the LPAR implementation is that it gives fine-grained allocation control over these individual resources, allowing them to be combined in almost any quantity and combination to create a logical partition.
Dynamic LPAR extends these capabilities by allowing this fine-grained resource allocation to occur not only when activating a logical partition, but also while the partitions are running. Individual processors, memory regions, and I/O adapter slots can be released into a "free pool", acquired from that free pool, or moved directly from one partition to another—again, in almost any quantity or combination.
Resources moved by dynamic LPAR operations have the same full set of capabilities that they would have if assigned to the partition at boot time. For example, a moved processor has full access to all of the partition's memory, I/O address space, and I/O interrupts, and so can participate fully in supporting that partition's workload.
Users of the affinity partition configuration option, which allocates CPU and memory resources in fixed patterns based on multi-chip module (MCM) boundaries, should note that only the I/O adapter resources can be dynamically reconfigured while in that mode.
While this introductory release of dynamic LPAR does provide the full capability of moving resources between running partitions, it does not mean that such LPAR resource movements will be occurring in a spontaneous or unexpected fashion. Rather, it means that LPAR resource movements are fully dynamic (performed non-disruptively while partitions continue to run) but not necessarily automated (driven independently by some internal policy or condition/response). As will be described later, the products are enabled so that scripts can be written to drive any sort of dynamic LPAR automation. This capability also enables future offerings in server automation and workload management.
|
Dynamic LPAR provides even more flexibility in dealing with changing workload demands and server deployments. Some obvious examples include:
- Move processors from a test partition to a production partition in periods of peak demand, then move them back again as demand decreases.
- Move memory to a partition that is doing excessive paging.
- Move an infrequently used I/O device between partitions, such as a CD-ROM for installations, or a tape drive for backups.
- Release a set of processor, memory, and I/O resources into the "free pool", so that a new partition can be created from those resources.
- Configure a set of minimal LPARs on a single system to act as "failover" backup servers to some primary servers, and also keep some set of resources free. If one of the associated primaries fails, then assign free resources to that backup LPAR so that it can pick up the workload.
As can be seen by these examples, dynamic LPAR opens a whole new set of possibilities for improving operational efficiency and getting more value from server hardware investments.
|
