IBM's recent announcement of its TotalStorage DS6000 and DS8000 storage arrays is a pretty good indicator that Big Blue is serious about refreshing its storage offerings. Especially interesting about this announcement is IBM's introduction of logical partitioning (LPAR) technology on some of its high-end DS8000 models.
EMC, Hitachi and IBM have been separating I/O workloads on their high-end devices for several years now, assigning I/O pathways to physical assets such as arrays, HBAs, ports and cache. Of course, even at the desktop level we have been physically partitioning individual disks since the 1980s. But logical partitioning is a whole new kettle of fish.
Physical devices, once their storage has been virtualized into a single large storage pool, can be partitioned and repartitioned logically. But why would you want to do this? There are several reasons, all of them good.
It is important to note that the LPARs allow each partition to exist independently of the others within the box. So as they like to say in Las Vegas, what happens there stays there.
Isolation such as this can be important when operations within a partition are potentially harmful to other data or processes running elsewhere. Extreme examples might the need to ensure that a security violation affecting one data set cannot have an effect on adjacent data, that the vagaries of an R&D system don't impact data servicing production systems, or the imperative that, should one part of a storage system crash, it doesn't bring everything else down with it.
A more mundane example might involve moving some planned process - necessary to the business but known to suck the cycles out of a CPU - to an isolated partition so that its operations don't degrade performance of other operations. Think of how your own operations might improve if you could replicate data to an isolated partition and then run your database reports or backups from the replica. Production systems would be untouched and business processes accessing the original data would not be degraded.
Of course, another way to isolate data is to physically split it off to other machines. A reasonably good analogy of this lies in a comparison of LPARs and storage-area networks (SAN).
SANs typically offer greatest value when the storage pool they represent is as large as possible, but in order to isolate problems and to keep them from cascading across the storage network many managers are forced to run multiple SANs. The good news is that the data is isolated. The bad news is that there are now two networks to manage, and the physical assets may be widely distributed.
In the future, IBM plans to add capability for running applications on an LPAR (requiring, naturally, that the application has been ported to IBM's Power processor and its AIX-based operating system.) Today, managers have the option of configuring the DS8000 as a monolithic storage pool or, using LPARs, as multiple storage systems. Using LPARs should add efficiency in several management areas, and is also likely to help rein in floor space, power and cooling requirements.