Tiered Access: iSCSI for the Enterprise, Part 1

iSCSI isn't just for small data centers. To understand how iSCSI fits in your data center, we look at the four important tiers of servers. (Part 1 of 2)

Internet Small Computer System Interface (iSCSI) has primarily been thought of as a way for entry-level or small data centers to consolidate storage. While iSCSI does just that, it should also be considered by medium- and large-sized data centers for developing a tiered access strategy. This two-part series discusses how enterprises can take advantage of tiered access to storage, which allows servers to access consolidated storage at performance levels and price points that are appropriate for: (1) a specific server, (2) the applications on that server, and (3) the storage attached to the storage infrastructure.

Tiered storage access for information/data lifecycle management (ILM) best fits in medium- to large-size data centers. In the smaller data center, an "iSCSI only" strategy is typically more cost effective.

As with ILM, the first step in implementing an "iSCSI only" tiered access strategy is to determine what performance levels the servers attaching to the consolidated storage system will require.

Four Tiers

In most enterprises, servers are classified by their level of importance to the business and by specific performance requirements.

In the top tier (Tier 1) are mission-critical servers. These servers require the highest levels of performance and reliability because they typically run ERP applications or provide real-time information. In most enterprises, however, mission-critical servers make up less than 15 percent of the server population. These servers will most often be attached to a fibre channel (FC) storage area network (SAN).

In the second tier are business-critical servers. Such servers are important to the business, but either do not require high disk I/O performance or can afford downtime of up to a few hours per year. These servers typically run departmental applications, file services, or content-storage applications such as document imaging.

Tier 3 is comprised of "Important" servers, where consolidation is more a matter of convenience for data backup and storage management, and for better utilization of the primary storage investment.

The fourth tier of access applies to archive servers; that is, servers used for accessing old files, archiving and for disk-to-disk backup.

Goal of Tiered Access

The goal of tiered access is to increase the number of servers that will attach to the SAN. The way to achieve that goal is to reduce the cost per attached server. iSCSI reduces costs over a traditional fibre channel SAN in two ways:

  • The cost of the iSCSI Host Bus Adaptor (HBA), the card that provides access from the server to the consolidated storage, is low. This card is similar in function to a SCSI card connecting a server to an external array, except iSCSI connects the server across an IP network to external storage. The cost of an iSCSI HBA is about $1,000 less than an equivalent fibre channel HBA.

  • A fibre channel storage network or SAN is built around FC switches costing approximately $1,000 per port. iSCSI servers, on the other hand, can be attached through existing 1 gigabit Ethernet (GbE) architecture at a cost of about $350 per port.

With these two factors combined, connecting a server to networked storage in a highly available configuration (redundant cards and data paths) can be accomplished for approximately $3,300 less using iSCSI. These savings enable an enterprise to connect more severs to the SAN, helping to increase the number of attached servers.

iSCSI for the Enterprise: HBAs

Once an access requirement is determined for each server, the appropriate HBA needs to be installed. This HBA will connect to the storage network. The server's access tier determines the type of HBA used.

Mission-critical servers—those of Tier 1 with either high-performance or high-availability requirements—will use a traditional FC HBA and will connect to FC ports on a storage switch with FC attached storage connected to it. In addition to performance and/or reliability, FC HBAs may also be selected for OS compatibility reasons. However, iSCSI does not yet cover all operating systems.

Servers in Tier 2 are the first to employ iSCSI. Servers in this tier require above-average performance and reliability, but not the cutting-edge performance of FC. The host bus adapter we suggest here is an iSCSI HBA. This adapter processes the iSCSI and IP protocols on the card itself, enabling an iSCSI host to achieve performance levels near the theoretical maximum of 1 GbE while at the same time reducing the load on the host server’s CPU to below 2 percent.

The motivation for consolidation of important servers—those in Tier 3—is more a matter of convenience than of necessity. It allows these servers to take advantage of data protection features, such as snapshots and remote replication that are included on most enterprise storage arrays. They may also benefit from using iSCSI software initiators on top of standard (and less expensive) 1GbE network interface cards (NIC). While this will increase the CPU load on these servers, typically they have abundant CPU cycles and are less disk I/O intensive.

The final tier, Tier 4 (archive servers), also can take advantage of software-based initiators and standard 1GbE cards. Since the data on these servers is old and is seldom accessed, both their disk I/O requirements and CPU utilization are so low that standard access is more than adequate. What makes this tier unique is that its access parameters may be too low to even require 1GbE cards. Many customers use 100BaseT with an iSCSI software initiator with these servers, further reducing costs.

Most of these third- and fourth-tier servers are excellent candidates for iSCSI. While they don’t need the "bleeding edge" performance of fibre channel, they can still benefit from the improved storage resource utilization and data protection capabilities that a consolidated storage solution provides. Performance can be enhanced through the use of redundant iSCSI HBAs in the host. Trunking of HBAs is supported, although failover of the NIC can take more than ten seconds.

Next week I'll discuss iSCSI as it relates to infrastructure. I’ll also address enterprise iSCSI vs. fibre channel, specifically addressing whether it's an “either/or” choice.

About the Author

George Crump is the vice president of technology solutions at SANZ Inc. gcrump@sanz.com