Enter the cube, exit the connector: IBM introduces a brick of a concept for storage.

In many of the industry's computing giants, it is easy to wonder how committed they may be to mass storage. They certainly acknowledge the importance of the storage element of computing, but the business dedication is more often given to servers and other processors. This somewhat flies in the face of recent research (BBC Consulting). But one computing giant that is giving time and resources to the next generation of storage problems is IBM. Of the eight major research operations that IBM maintains, three facilities and upwards of 140 scientists are seeking to solve pain points for users. A site visit to IBM's Almaden Research Center highlighted some of these activities.

Challenging Applications Contention

According to IBM Fellow Jai Menon, one of these activities focuses on issues of block virtualization and controller futures. Virtualization, one of the most used and sometimes abused words in the storage vocabulary, is a popular storage management strategy hiding elements of management complexity from end users and treating disparate storage elements as a pooled resource. But the masking of the complexity and pooling do not at this time address the problems of applications contention on the bus.

To address the problem, IBM is working on a virtualization engine that focuses on very low latencies through the virtualization layer, a scalability capability to the hundreds of terabytes in a single SAN-wide pooi, N-way coherent read/write caching for up to N x 32, and the use of what Menon called "performance assured LUNs" (Figure 1).

The performance assured LUN is a project to create a LUN with 100GB capacity and 10 msec response time at 10,000 ops/sec. Quality of service per LUN will be achieved through the placement of LUNs, the assignment of LUNs to the virtualization engines, I/O scheduling and cache management. The capacity, response time, and operations per second should help with applications contention as well as total cost of ownership.

IBM's research is not confined to the block level. They are taking steps toward the next generation of common file services and management as well. For example, currently, free space in one operating system is not usable by another OS. The next generation will give free space under one OS to another as needed. Further, the emerging file services will allow file sharing, the ability to move applications between operating systems (without moving data) and the achievement of a single name space.

Virtualization in Action

Some of the work that IBM has been doing in the virtualization space has been realized in the reintroduction of the IBM iSeries servers. The new IBM eServer iSeries allows customers to add new capacity on the fly, run up to 10 Linux servers on a single processor, and install IBM's ebusiness software for the price of a competitive UNIX system.

The relaunch is part of a two-year, $500 million initiative to re-energize the IBM eServer iSeries. The new systems include IBM's leading-edge hardware and software technology, run multiple operating systems including Linux and OS/400, and allows small and medium businesses the ability to consolidate and manage Windows applications and data, reducing cost and complexity.

The introduction included an On/Off Capacity Upgrade on Demand. The on/off capacity upgrade on demand allows mid-market customers to pay for extra server resources by the day, without taking the server down. Customers can now choose from Temporary or Permanent Capacity Upgrade on Demand options.

Storage Tank Approaches

IBM is putting a lot of eggs in one storage management basket. Called Storage Tank, it is scheduled for release during 2003. IBM has great hopes for Storage Tank, hoping the file system will simplify the management of data sitting on servers and storage systems made by a wide array of vendors (Figure 2). The Storage Tank file system should give administrators a way to pool servers and storage hardware and manage the data used by these systems from one central location. The technology would help remove barriers caused by different vendors' hardware and allow data to travel across various file systems without losing management policies set by the administrator.

Storage Tank is designed for hardware from IBM, Sun Storage, HP, and others in a common SAN (storage area network) to share the same view. Other companies that have been trying to develop that type of technology have been smaller startups for the most part, but IBM could very likely engineer a success if it can convince other top-flight storage OEMs to make common cause with them.

Like the virtualization engine, Storage Tank is designed to improve capacity use through central management of all file system data and unify it with the block level data on the SAN. It will also permit end-users to assign classes of storage, which empowers those users to make better acquisidons plans. The software features storage consolidation capabilities and a single name space for easier management. All of this, of course, contributes to a reduced cost of ownership, which is the expression of the end user's pain or the lack thereof.

Autonomic Storage

The term "autonomic" as applied to storage is so frequently used that even the IBM research scientists wince at its use. But the fact remains that the total cost of storage continues to increase due to the cost

of storage servers, the rising complexity of storage management and the pervasive need for 24x7 operations. Autonomic storage is aimed at reducing the total cost of storage by providing systems that are largely self-managing, self-diagnostic and transparent to the user.

The Storage Systems group at IBM Almaden Research Center currently has several research projects in progress to address the complexities of autonomic storage. A key project in this area is that of "Collective Intelligent Bricks" (CIB for short, Figure 3).

Formerly known as Project Ice Cube, the CIB is the concept of using simple storage "bricks," each of which contains a microprocessor, a small number of disk drives (about 12), and network communications hardware to provide a data storage system that scales to petabytes of storage. Such a storage system will increase the reliability and reduce the size, cost and power usage of the vast storage systems needed in the future.

The bricks are perfect cubes that stack in a small footprint, resembling sugar cubes or ice cubes (hence the code name). The bricks are attached on vertical columns for power insertion and heat removal; a brick communicates with adjacent bricks in a 3D mesh.

Exterior connectors, fans, cables, fibers or lasers are not used, allowing this tight stacking of modules. Instead, connectivity between the bricks is accomplished using a capacitive coupler. Rather than connecting through pin-in-socket connectors, the capacitive coupler aligns film elements tightly squeezed against another. The film conducts the energy and becomes something of a capacitor to one another. The goal for this connectivity strategy is 10 Gbit/sec at press time.

Planting a Flag

After your author's visit to Almaden Research Center, there is no doubt that IBM is planting a flag to claim technological dominance in the mass storage space, both in hardware and software. If everything goes according to plan, IBM will be able to offer a common enterprise file system for all servers and a virtualization engine that will pool all storage arrays. These plans are congruent with such industry trends as SAN/NAS convergence, open SAN management, and the location of intelligence in the storage subsystem.

Coping with spiraling storage capacities, both in terms of hardware and software tools, is a nontrivial task, and it's only becoming more difficult. IBM strategies bid fair to reduce the pain that enterprise and mid-range IT managers feel. How will he rest of the industry respond? That is another story.