iSCSI: changing the storage landscape.

In today's ever-expanding business model, more and more information and applications are being housed outside the corporate data center as companies struggle to store, protect and provide access to an ever-increasing amount of data throughout the enterprise. While direct attached storage (DAS) has been the dominant storage model for servers, there has recently been a strong movement toward networked storage technologies (such as the new standards-based Internet Small Computer System Interface (iSCSI) technology), which have the potential to significantly increase the proliferation of affordable networked storage.

Companies that have had a DAS infrastructure for many years are discovering that these legacy systems have become a business issue. While older servers' computing capability might be adequate, their storage subsystems may be obsolete, due to lack of expandability or aging reliability. While DAS often works well in environments with a limited number of servers, significant data growth can quickly make the situation unmanageable. The only way to accommodate growth is to add servers; yet since each server must be managed individually, every additional server increases management expense and difficulty.

The inherent inflexibility of DAS makes it difficult to deal with unplanned growth; it is tough to scale, and unused resources cannot be redeployed. This often leads to under-utilized, over-provisioned resources, leaving departments and enterprises with large amounts of unused capacity that might be needed someday.

While traditional Fibre Channel storage area networks (SANs) overcome a number of the issues posed by DAS, they are expensive to implement and maintain. Network attached storage (NAS) has traditionally provided cost-effective file data storage, but has limited applicability in applications--such as databases--where DAS typically resides. However, the emergence of standards-based iSCSI technology offers the benefits of a SAN with the cost-effectiveness of NAS over an existing Gigabit Ethernet IP network. The cost, flexibility and efficiency of this new class of IP SANs offer some compelling business reasons for moving to this storage architecture.

The creation of an IP SAN requires host application servers equipped with an iSCSI initiator driver; a standard Gigabit Ethernet NIC, TOE card, or iSCSI initiator HBA; an Ethernet switch; and an iSCSI storage target. The iSCSI protocol is a client/server model, whereby an iSCSI initiator initiates requests and an iSCSI target receives the requests and delivers the requested data. The iSCSI stack at both ends encapsulates SCSI commands and block data for transmission over TCP/IP.

By separating the storage from the server, an IP SAN allows flexible storage to reside anywhere on the network and, under certain circumstances, can provide access to that block-level data from multiple servers.

In traditional storage scenarios, different storage devices had to be deployed for different access topologies: one for corporate applications that require access to block data, and another for file access using CIFS/NFS network file-sharing protocols. With iSCSI's ability to transfer block data over an IP network, both files and block data can be centralized on a single storage server. This "unified block and file" architecture provides the benefit of SAN-style block-level storage at speeds that are acceptable in many applications, and greater reliability, at a significantly lower cost.

Physically consolidating storage in an IP SAN centralizes storage management, making it possible to manage one storage server or every storage server on the network from a single location. IP SAN storage devices can be increased with expansion disk arrays and the capacity of existing volumes can be expanded, usually without requiring downtime or RAID rebuilds. Storage can be provisioned as needed.

In terms of availability, iSCSI storage volumes behave like dedicated local storage with underlying RAID for added data protection and redundant hardware to minimize downtime.

IP SANs:

* Are easy to use, deploy, and understand

* Improve storage utilization, scalability, and availability

* Connect with standard Ethernet hardware and existing plumbing

* Have the lowest initial investment of any SAN solution

* Offer the lowest TCO of available storage topologies

Whether replacing a legacy DAS system or as an adjunct expanding the capacity of existing servers, an IP SAN built on iSCSI standards can provide advantages to many applications throughout an enterprise. Examples include:

* Saving Microsoft VSS shadow copies on an IP SAN instead of on the Windows server provides fast, user-accessible file/folder recovery, and easy access to previous versions while reducing demand on IT.

* For Microsoft Exchange archive applications, an IP SAN provides fast, economical block storage and recovery without the cost and complexity of clustered or proprietary solutions.

* Storage consolidation using IP SAN can extend the life of aging servers by adding capacity and providing extremely cost-effective and scalable block storage for back office, small and medium business, retail operations, and departments.

* IP SAN is ideal for nearline archiving for any application that requires a common storage pool. The shared storage pool provides a solution that is affordable, easily managed, and easily scaled to meet growing block storage needs.

* An IP SAN with two iSCSI targets and RAID-1 mirroring at the host can be used for any direct attached storage application with cost sensitivities and data redundancy requirements. RAID-1 has minimal overhead on the host resources, yet provides simple, easily implemented and cost-effective redundancy for failover, and protects against a complete failure at either of mirrored storage targets.

Things to consider when planning an IP SAN for an application include:

iSCSI initiator: Initiators range from a free driver that uses the CPU for network/iSCSI stack processing, with generally good results, depending on the application resource needs, to hardware initiators that start at $525 and offload cycles from the CPU.

When a software initiator is used, the iSCSI driver communicates through a standard Ethernet NIC. The use of a TOE card depends mainly on the relative horsepower of the CPU(s) to handle the application and processing of the network stack. The hardware initiator provides maximum benefit to the host CPU processing by offloading both the networking and iSCSI stack processing, and also provides remote booting capability.

iSCSI target device: The key thing about an iSCSI target device, in addition to support for the chosen initiator, is capacity. It should be able to handle current capacity needs and provide scalability for growth. iSCSI targets with dynamic capacity provisioning and expansion are preferred since expansion does not affect system availability.

Security: Associating specific LUN on an iSCSI target with specific hosts can be managed through a few iSCSI access methods, such as Challenge-Handshake Authentication Protocol (CHAP), which uses an authentication token to authorize access to a particular initiator. Also available is iSCSI Naming Services (iSNS), which sets up a service in the environment to authorize initiators and targets to automatically connect with each other. In addition, some iSCSI targets provide a form of LUN masking to define, at the iSCSI target, which host initiators can connect. Unfortunately, some software initiators are still immature in this regard and try to attach to any iSCSI LUN it finds. Using CHAP, iSNS, or iSCSI LUN masking will help avoid this.

Measuring the potential Return on Investment (ROI) of an IP SAN is similar to measuring the OT for any storage. There are two types of costs associated with purchasing new storage, no matter what interface is used. The first and most easily recognizable are the hard, or direct, costs that bear on the IT budget. Typically, these are out-of-pocket expenses at the time of purchase, and include capital equipment spending, labor, outsourcing, professional services, installation, support contracts, and training. The second type of cost is soft, or indirect. These costs are often felt, but not measured. Less tangible than hard costs, they typically result from reduced productivity or lost revenue when a system goes off line (such as for upgrade, repairs, or to accommodate large backup windows). Both hard and soft costs must be considered when calculating TCO for any storage investment.

Factors most relevant to the IT team's evaluation include:

Hardware: Installation, support, scalability

Software: Administrative (discovery and configuration), data management, reporting (usage and failures)

Finally, determine and quantify the benefits and risks associated with the IP SAN:

* Will current availability needs be adequate for a consolidated single-storage pool with RAID protection and redundant hardware?

* Will iSCSI performance (measured to about 80% of FC performance) be adequate for business applications throughout the enterprise?

* How will the organization gain from IT spending less on hard costs, and less time in storage management?

* What is the anticipated data growth over the next three years and the costs associated with meeting this capacity need with DAS or with IP SANs?

Many organizations may find that switching to an IP SAN provides significant advantages for data storage and access in terms of costs, manageability, reliability and productivity.

Steve Rogers is director of technical marketing at Snap Appliance, a division of Adaptec (Milpitas, CA)

www.adaptec.com