3.6 Final Project Definition
Use the information from assessments of the system and application I/O characteristics to define the project parameters. The type of SAN and the I/O behaviors point to the performance parameters and host system behavior expectations. The project definition also takes into account failure modes and other operational considerations such as dynamic SAN reconfiguration. Use the definition as a yardstick for measuring whether or not the goals of the SAN have been accomplished.
NAS Replacement SAN Definition
The definition of the design for the NAS replacement SAN is fairly simple. (See Figure 3.1 on page 59.) The parameters that drive the design are the bandwidth required for the application and multipath I/O channel infrastructure that prevents a systems outage in the case of a single I/O channel failure.
The bandwidth required is minimal, with a peak measured usage of 13MBps. This means that any single Ultra SCSI or Fibre Channel I/O interface can meet the bandwidth requirement for this SAN. The multipath I/O channel configuration requires a minimum of two channels per host system or storage system. Because two I/O channels provide from 72MBps to 200MBps, depending on the selected type, the bandwidth requirement can easily be met. The SAN requires 1TB of storage to accommodate its current data set and an additional 0.5TB of storage to accommodate six months of growth. All of the interconnections between fabric devices, if any are necessary, will also require two I/O channels.
Storage Consolidation SAN Definition
The definition of the storage consolidation SAN project is more complicated due to higher performance requirements and more tradeoffs to accommodate the different host systems and applications. (See Figure 3.4 on page 63.) Fabric bandwidth is one of the defining parameters of the SAN. Although only one of the systems has bandwidth requirements in even the average range for a single host system, the bandwidth requirements of all the systems being consolidated must be serviced concurrently on the SAN fabric. The storage consolidation SAN requires a multipath I/O channel configuration for failure resilience and load balancing, if possible. This SAN supports a data warehouse ETL workload, so the SAN includes a data movement tool that improves data transfer times and removes load from the consolidated host system's IP networks.
Aggregate bandwidths of 400MBps in the fabric and 100MBps per host system are necessary in this SAN. This performance should be adequate given a more evenly balanced workload across all of the systems. A balanced workload eliminates the spikes in the peak usage of the one host system with needs that exceed 100MBps. The SAN requires at least a pair of Ultra SCSI II controllers (or faster) to meet the SAN host system performance requirements. Because of the multiple controllers required for bandwidth, the multiple channel I/O failover and load balancing configuration requirement can also be met. The storage space required for this SAN is 4.5TB at the start.
This allocation provides storage space for the current data set on all three host systems, plus the capability to grow all three host systems by 0.5TB or any individual host system by up to 1.5TB on an immediate need basis.
Capacity-Planning SAN Definition
The bandwidth and flexibility requirements of the host systems characterize the project definition for the data warehouse capacity-planning SAN. The requirements also include a multiple I/O channel configuration for host system and storage device resilience. Features include data replication for scalability and disaster recovery that support the business-critical data warehouses targeted for the SAN.
Each host system requires 200MBps of bandwidth for storage devices, and the fabric must support the aggregate traffic of four host systems. These requirements mean that the fabric will require 800MBps of bandwidth to support the concurrent load of the host systems. The storage devices must also support the 200MBps from each host system either individually or as a group, depending on their size and the final allocation to each system. Two Fibre Channel I/O channels can meet the bandwidth and multiple I/O channel failover needs of each system. Only two I/O channels require high per-channel IOPS performance, so a trade-off that installs more I/O controllers to meet the IOPS needs of the host systems may be necessary. A higher number of the same, or lower, performance I/O channels can meet the IOPS needs of the host systems and provide a lower per-channel IOPS solution. However, the lower performance I/O channels might not meet the bandwidth needs.
The storage space requirement for the capacity-planning SAN is 1TB per deployed system or 4TB total to start. It is likely that there will be data growth, so some expansion capacity can be built into the SAN. To provide for the data replication scheme, the SAN requires installation of some additional fabric connectivity in order to increase available bandwidth without slowing the data warehouse application systems usage. Chapter 4 shows how the host design parameters defined here can translate into useful SAN designs.
Other SAN Types
The SAN design definition for a new project is set to meet the requirements of the project. A good strategy for setting these requirements involves finding applications or host systems that may have performance and host system needs that meet the requirements of the new project. Then apply the parameters of those systems to the new project SAN.
The design for an experimental SAN meets the testing requirements of the SAN. For example, if performing IOPS-limit evaluations, then use a low number of channels and a high IOPScapable storage device. If testing failover under stress, then specify at least one alternate I/O channel. Test SAN limits and behaviors by constraining the I/O parameter to be tested and then observing what happens to the host systems, storage devices, and fabric devices when an extreme load is placed on the SAN.