Reducing the cost of storage management.

Manageability is a top concern among IT and data center managers, for good reason: integration, deployment and operating costs can be from five to eight times the cost of hardware and software acquisition (Figure 1). Manageability involves a variety of key activities and assets such as installation, configuration, set up, asset tracking, firmware and operating system upgrades, identifying and isolating faulty servers, performing backup and restores, and managing mirror and data duplication servers.

How can manageability be improved and management costs reduced? Server and storage management can be improved at several different levels from component to server to data center. This article looks at ways of reducing storage management costs at each of these levels.

Component Integration

Increased component integration at the silicon level improves overall reliability and density of servers and storage boxes, and enables smaller form factors. Using System-on-Chip IO Processors (IOP) in Host Bus Adapters (HBA), servers and external storage boxes can significantly reduce failure rates and improve mean time between failures.

IO Processors typically include integration of:

* Core processor/s: RAID processing can be offloaded to a core processor, thus reducing the burden on the host processor and improving overall performance

* Memory controller: The IOP typically integrates a controller to maintain a large cache memory, thereby hiding latency to and from the storage disks

* PCI-X interfaces: These Peripheral Component Interconnect interfaces can be any bus interconnects. One interface is used to connect to the host bus, either initiator or target, and the other provides the capability to use standard Network Interface Controller (NIC) components

* Cyclical Redundancy Check (CRC) engine: The iSCSI specification includes a special calculation to help assure data integrity. When not performed in hardware, this calculation can be a significant burden to software and the host processor

* SATA/SAS controller: Offers low-cost RAID solutions. Drives can be directly attached to the HBA with no additional logic or controllers needed.

* XOR engine: iSCSI targets typically use RAID technology to maximize data integrity at the disks, and nearly all high performing RAID solutions use XOR in their implementations. Since XOR operations can be a large burden to software and the host processor, they are better performed in hardware.

IOPs facilitate development of single chip RAID HBA, ROMB (RAID on Motherboard), and thin form factor external storage boxes. ROMB solutions can increase the robustness and reliability in Direct Attached Storage (DAS) requiring no additional HBA or controller hardware. Typically, different RAID configurations such as RAID 0, 1, 2, 3 and 5 are supported. External storage boxes built around System-on-Chip (SoC) devices have small form factors and are designed to be cost effective.

[FIGURE 2 OMITTED]

Storage Area Networks (SANs)

In recent years, the need to organize increasingly vast amounts of storage has created demand for a robust, scalable and manageable external storage solution. The storage area network (SAN) was designed to meet these needs. A SAN offers IT managers the ability to move storage onto a storage-specific network, where it can be consolidated and managed with tools and processes specifically designed for this purpose. Today, a majority of SANs deployed are based on Fibre Channel (FC). However, Internet SCSI (iSCSI) is emerging as a strong and low cost alternative. Figure 2 shows an iSCSI-based SAN implementation.

Both FC and iSCSI based SANs provide many of the same benefits:

* Centralized management: Promotes more efficient management and encourages sharing of documents rather than maintaining multiple duplicate copies

* More effective use of storage: By allowing for storage consolidation, these technologies typically mean fewer servers with unused capacity and fewer servers at full capacity. Instead, the servers share a pool of storage

* Higher availability: Through the use of redundancy in both SAN infra-structure components and storage components, SANs provide for higher availability than a simple server with direct-attached storage

* Faster backup and restore time: Backups can now take place on isolated networks, and since SAN traffic is block-level, data can be backed up faster than with traditional file-based backups

* Easier deployment of new capacity or load balancing of existing capacity: SANs allow for IT managers to add capacity to a single infrastructure and make it available to one or more servers without opening the server. Advanced features allow for more dynamic capabilities, including the transparent addition, deletion or movement of storage resources from one server to another

* Enabling new capabilities such as remote mirroring: Remote mirroring, snapshot copying and storage virtualization are all examples of capabilities that are enabled by the SAN

Benefits and Uses of iSCSI

Although they provide valuable benefits, Fibre Channel SANs can be fairly expensive. They require specialized equipment not generally present in an organization's network environment and skills that often aren't available among personnel. By contrast, SANs based on the iSCSI protocol leverage the Ethernet network infrastructure and the skill sets of IT personnel.

The iSCSI protocol makes it possible to encapsulate traditional SCSI commands (also used by Fibre Channel) into TCP/IP packets. This is what enables iSCSI SANs to use traditional LAN network infrastructure components such as hubs, switches and routers. IT personnel already have the necessary training and logistics for the existing infrastructure, so it usually takes minimum effort to support an iSCSI SAN as well. iSCSI also enables quick adoption by allowing either pure software implementations or accelerated hardware solutions. The future of iSCSI is expected to evolve as Ethernet evolves, with continuing advances in performance, security, reliability and scalability.

ISCSI opens up the benefits of consolidated storage not only to different kinds of companies, but to companies of many sizes and budgets. These include:

* Internet service providers (ISP)

* Storage service providers (SSP)

* Organizations that need remote data replication and disaster recovery. Using an iSCSI SAN, all of its critical data can be located on one SAN, and capabilities such as remote mirroring can be used to leverage cross-country IP infrastructure and provide for real-time data backup thousands of miles away.

* Small office/home office businesses with limited IT resources, infrastructure and budget. ISCSI allows these organizations to reap the same benefits at larger companies with more resources and that had previously deployed FC SANs

Modular Computing

Adoption and migration to modular or blade servers from pedestal and rack-mounted servers is another approach to simplifying server and storage management. Blade servers form the hardware foundation for a modular computing paradigm. Modular computing can reduce complexity and enable greater automation and virtualization of resources, creating reliable systems with essentially no single point of failure.

Modular computing integrates chassis-based modular hardware resources that are easily removed and replaced. These systems aggregate interconnect, cabling, switching, power supplies, cooling, and other resources, simplifying the infrastructure and its management and service. Hot pluggable modules including compute and switch modules allow easy serviceability.

In typical blade server products, the same blade server chassis has integrated network switch blade/s and storage blades (FC blades). By integrating Ethernet and FC switches into the blade chassis, cabling and management and service workload is reduced. Having FC switch blades enables seamless integration with the existing FC SAN infrastructure. Ethernet switch blades allow integration of existing Network Attached Storage (NAS) infrastructure, iSCSI storage and Ethernet networking infrastructure.

By reducing complexity, blade servers can simplify service requirements. Blades share chassis-based resources, including redundant power supplies, redundant management and network access, and communicate over redundant interconnecting back or mid-planes. For example, a rack-mounted server (RMS) configuration with 14 servers can require 112 cables, while a fully populated, 14-blade server chassis needs as few as eight cables (this example is based on Intel's modular server). To service a modular server system, the blade or modular component is simply removed and replaced; modules are hot plugged and no special tools are required.

Software Deployment Managers

Still another way to reduce the overall cost of server and storage management is to use software-based, network centric deployment managers. Deployment managers simplify and automate the process of deploying, provisioning, and managing a heterogeneous mix of servers and storage boxes in a datacenter. An administrator can manage a number of systems from single console, typically using a Web interface.

Deployment managers allow for inventory management, tracking software licenses, setting up administrative and user access, and setting up server specific customized environments. There are several OEM and third-party deployment managers available in the market, including OpForce from Veritas, Platespin's Operations Management Center, Altiris' Management Suite, and IBM's Remote Deployment Manager (RDM).

Remote management and operation is made possible by the use of standard firmware interfaces. Protocols such as BOOTP, PXE (Pre-Execution Environment) and TFTP (Trivial File Transfer Protocol) allow server and storage deployment remotely. A storage server connected to a network can advertise itself to the deployment manager or provisioning server.

Servers built with alerting mechanisms provide timely alerts to the data center operators. Network-based management software solutions have provision for detection and isolation of faults. When a fault is detected, failed servers can be isolated and failover paths established. At a server level, protocols like Intelligent Platform Management Interface (IPMI) support failure detection and alerting.

IPMI defines common interfaces to "intelligent" hardware used to monitor a server's physical health characteristics, such as temperature, voltage, fans, power supplies and chassis. These capabilities provide information that enables system management, recovery, and asset tracking. IPMI includes capabilities to automatically alert IT managers of system errors and enable immediate system recovery.

In addition, security and authentication management is simplified as well as the ability to launch virus patches to a range of servers with ease.

Storage Management Initiative (SMI)

Managing a heterogeneous data center environment with servers and storage entities from different vendors is still a complex problem. Integration of different management tools to perform basic storage discovery, topology mapping and (potentially) provisioning requires a lot of effort and fine tuning. This problem is not lost on manufacturers and vendors.

The Storage Networking Industry Association (SNIA) has undertaken the Storage Management Initiative (SMI) to address these specific issues in the context of Storage Area Networks (SAN). The published aim of SMI is to develop and standardize interoperable storage management interfaces that allow different classes of hardware and software products supplied by multiple vendors to reliably and seamlessly interoperate for monitoring and controlling resources. This specification is called SMI-S (SMI-Specification). It defines an open system standard application program interfaces (APIs), allowing applications to manage storage devices manufactured by various vendors.

SMI-S standard provides for heterogeneous, functionally rich, reliable and secure monitoring/control of mission-critical global resources in complex and broadly distributed multi-vendor SAN topologies. Recently, the SMI-S 1.0.1 specification has been released. SMI-S 1.0.1 represents the first step of several planned iterations of the specification, steps that will go beyond standardizing management of SANs to include NAS, iSCSI, and other storage networking technologies. Most of the leading vendors of hardware and software have committed to supporting this emerging standard and products are expected to roll out starting in 2004.

Storage management can be improved at a variety of levels from silicon to server design to the data center and beyond. Component integration, SANs, iSCSI technology, modular computing, software deployment managers, and the Storage Management Initiative all offer ways in which the management costs of storage can ultimately be reduced.

Hardware & Software       $1.0
Utilities                 $0.5
Operations                $2.7
Integration & Deployment  $2.3

Source: Intel Corp. data based on Gartner Group algorithms, 2003

Figure 1: Operating and maintenance costs are much higher than the cost
of HW and SW.

E. P. Komarla is engineering manager, Storage Components Division, Intel Corporation (Santa Clara, CA)

www.intel.com

www.altiris.com

www.ibm.com

www.platespin.com

www.veritas.com