Available Alternatives | Storage Networks: The Complete Reference

Considering all the available estimates, and validating the capacities , processing requirements, and services levels, the new data warehouse infrastructure would need to support a 3-terabyte capacity. Further investigation indicated that the user community-although comparatively small, with less than 250 end users-would require an I/O workload of multiple gigabytes per transaction. This also carried over into I/O workload for loading and updating the database, creating an additional multiple-gigabyte workload per transaction. In summary, calculations indicated a total I/O workload of 29,300GB for a 24-hour period. This was reduced to 1200GB per hour and subsequently an aggregate of 20GB per second.

Using an external direct attached storage strategy that would simply enhance the current configurations would require additional servers and additional SCSI storage arrays. A quick estimate revealed a requirement of approximately 50 new servers with the maximum of SCSI adapters for each server to meet the aggregate I/O workload of 20GB per second. An alternative to this scenario would be to replace existing servers with a much larger server machine, such as a UNIX mainframe server, where the machine could be partitioned to run the individual data warehouses with dedicated resources. However, this solution also would require the new storage arrays, although the existing arrays could possibly be used in conjunction with these.

By utilizing the guidelines for estimating SAN capacity (see Chapter 18), it turns out that a SAN configuration will require a total of 240 ports, including calculations for recovery and redundancy factors. Given the type of workload-datacentric (high I/O content), transactional with batch updates, and less than 250 end users-the configuration could be handled by 15, 16-port switches, or 4, 64-port director class switches. However, this solution also requires that new storage arrays be purchased for compatibility and performance within the storage network. This strategy does provide the possibility of utilizing the existing servers.

The storage team, in conjunction with technical support management, chose to evaluate the UNIX mainframe alternative compared with the SAN configuration. This evaluation took place as a mini-capacity planning exercise, since this requirement was addressed from an existing capacity plan that was in place. Consequently, the IT executives, to provide the flexibility to evaluate solutions outside the context of existing budget plans, invoked the end-user non-aggression pact. (Refer to Chapter 23 regarding storage network capacity planning and Chapter 22 for end-user non-aggression pacts.) This exercise revealed an interesting comparison.

Storage Evaluation Results

The UNIX mainframe was a high-end SMP machine that was capable of being partitioned into 2 to 16 logical processing partitions. The SAN configuration was analyzed as 4, 64-port director class switches, with 12 HBAs installed across the existing data warehouse UNIX servers-for example, 4 HBAs each. The storage arrays were configured with 3 terabytes of capacity with specific configurations for each solution-for example, one set for the SCSI-based UNIX mainframe and a different set of arrays for the Fibre Channel-based SAN.

The requirements and evaluation is further depicted in Table B-2. Our evaluation is based on a score of 1 through 10, with 10 being the highest, or best, score. Vendor selection is specific to the customer and the competitive nature of an acquisition. It is recommended that a minimum of three bids be requested once workload estimates have been concluded.

Table B-2: Storage Solution Evaluation Table for Data Warehouse/Data Mart Applications
Requirement	SAN	Direct Attached
Meets I/O workload performance and capacity specs	9	8
Ease of installation and maintenance	3	4
Scalability of storage capacity	7	4
Compatibility with existing storage infrastructure	5	8
Composite score	24	24
System price	$700,000	$1,850,000

Note	We have estimated pricing only as a general reference point. Included in these pricing estimates are total components for system operation, additional software required, and storage array costs.