HP ProLiant 4500 Disk Subsystem Performance and Scalability - Page 25

WHITE PAPER (cont.) Note: Within each test comparison, the total disk capacity remained constant. ECG025.0997 ... Multiple Disk Controller Test Results In our multiple disk controller tests we found that concurrency coupled with adding disk controllers to an environment increases system performance. For instance, Figure 13 displays a comparison between two tests in a RAID 5 environment. First, we tested twelve 4GB drives (11 data, 1 parity) using one controller. Next using two controllers, we equally split the number of drives (5 data, 1 parity) on each controller, totaling 10 data and 2 parity drives. Even though we had two drives dedicated to parity in the two controller environment, test results still show a significant benefit, yielding a 57% increase in system performance. 20,000,000 18,000,000 16,000,000 Disk Controller Scaling (RAID 5) 10+2 x 4GB, 2 Ctr 11+1 x 4GB, 1 Ctr Server Throughput (Bytes/sec) 14,000,000 12,000,000 10,000,000 8,000,000 6,000,000 4,000,000 57% 2,000,000 0 4 8 16 20 24 28 32 36 40 44 48 52 56 57 Number of Clients Figure 13: Disk Controller Scaling in a RAID 5 Environment. The disk controller scaling tests configured with no fault tolerance or RAID 0, shown in Figure 14, receives only a minimum performance gain of 3% between one and two controllers. Since we enhanced the disk subsystem and did not see any significant improvement, other factors must be limiting our throughput. Upon examination of the other subsystems, we found the processors nearly saturated. To remove this bottleneck, we would have to use faster processors and rerun the test. Nevertheless, remember that changing the disk subsystem enhances performance only if it is the bottleneck. In this case, the processors are the bottleneck and not the disk subsystem, so changing the disk subsystem provides no benefit towards improving performance. 25