|and DS8000 - Pricing|
Power of the DS8000
IBM hasn't just swiped some CPUs and a few spare server parts to tweak the DS boxes. The DS8000 incorporates two full pSeries P5 servers, featuring dual Power5 processors that offer high-end performance, larger caches, improved bandwidth and advanced features such as partitioning. The DS8000 scales linearly with the addition of processor pairs. Rich Lechner, IBM's VPof storage, calls this type of scalability a "new infinity" that will eventually allow the DS8000 togrow to more than 1PB of storage.
The partitioning capabilities of the Power5 processor are a key part of the DS8000 picture. Logical partitioning (LPAR) is built into the hardware and firmware, and can effectively control and allocate all system resources. Initially, two partitions are possible on the four-way model. Partitioning isn't just a matter of allocating storage and cache; each partition gets its own dedicated processors, cache, storage and adapter slots. With HDS' new TagmaStore, virtual machines can be similarly created, but with TagmaStore, "you can't do anything as far as allocating the processing power," says Hurley. Comparing IBM's and HDS' implementations, Mike Kahn, managing director of The Clipper Group, says, "The customer is presented with two different approaches, with different scaling and economics."
In the initial configuration that allows two partitions, there's a 50-50 split of processing power where each partition also has a dedicated bus, the RIO2--or Remote Input/Output 2--yet another element in the Power5 server architecture. Soon, IBM will offer the capability to partition on a 75-25 split, in which each side gets a dedicated RIO2 bus. "At this stage of the technology, we can't share the RIO interconnect between LPARs," notes Hartung. "In a future release, we will." He indicates that in future revisions, partitioning can be a dynamic process in which policies control resource allocation.
In IBM's server world, partitioning can be done on an even more granular level, with Power5 configurations divvied up into chunks as small as 10%. While redundancy and high-availability demands make it unlikely DS8000s could be sliced into such small segments, IBM has indicated that additional levels of partitioning are in the offing. When that happens, "you're going to be able to start tiering performance within the box," says ESG's Hurley.
For example, a single DS8000 can be partitioned to effectively operate as two separate arrays--to isolate a production environment from development storage, for example. Or one partition could be dedicated to open systems, while the other handles mainframe storage.
UPMC's Furmanski likes that IBM is applying some of its newest server technology to the DS8000 and says, "we've been waiting for it for quite some time." He sees possibilities for the DS8000's partitioning capability in his shop, which has about 140TB of storage on a mix of Hewlett-Packard (HP) Co. boxes and a few IBM Shark F20s. "We have to control and segregate the performance we get for our production environments vs. our development environments," says Furmanski, who envisions using LPARs to ensure the availability of his firm's electronic health-record system.
But slicing an array in half isn't the only application for LPARs. LPARs are independent but can interoperate, so a partition could run a storage management program or some other application. The benefit would be that the two partitions--the application and the storage itself--are closely coupled and would communicate at memory speed rather than across the network. "We're looking for applications that are going to benefit from co-location with the data," says BCBS' Venable, "and TSM and SVC are two primary workloads that we think will be good."
Furmanski is also intrigued by the prospect of running an application in a partition, notably SVC. "When you link it up with their storage area network (SAN) virtualization software and host virtualization software," he says, "they have an orchestrated environment of virtualization."
Another application that would take advantage of the DS8000's partitioning would involve running a NAS head in one partition, making it possible to have file and block systems in a single array, a capability NetApp has offered for some time. "The model is that of storage-based middleware," says IBM's Hartung, meaning the NAS file system runs within the array, rather than on an appliance.
Partitioning and the ability to allocate essentially all system resources puts the DS8000 in a unique position. The closest in concept is HP's StorageWorks Grid design in which "cells" containing processors, cache and storage have their own "personalities" (such as a NAS file or block file system), and can be added to the grid incrementally. "HP and IBM are the only ones talking about not just partitioning off capacity, but being able to partition off the computing power," notes Hurley.
This was first published in November 2004