Page 1:The 2011 Extreme RAID Project
Page 2:The SSDs: 16 x Samsung 470 (256 GB)
Page 3:The Controllers: 1 x LSI MegaRAID 9280-24i4e And 2 x LSI MegaRAID 9260-8i
Page 4:RAID Creation In Windows
Page 5:Benchmark Results: Throughput
Page 6:Benchmark Results: I/O Performance
Page 7:Benchmark Results: 4 KB Random Reads/Writes
Page 8:Conclusion: Second-Generation 6 Gb/s Systems Needed
Benchmark Results: 4 KB Random Reads/Writes
The higher the queue depth, the better the system performs. When we look at 4 KB random reads/writes, we see a different picture than from the Iometer benchmark. This is especially true in reads, where the dual-controller system achieves an excellent result of more than 300 000 IOPS at a queue depth of 32. The limited single-controller configurations are about half as fast. The curve shows clearly just how great an effect queue depth has on I/O operations. At QD=1, performance is below 30 000 IOPS.
In random writes, the situation is a bit different: low queue depths of up to QD=2 get the best results on the test system. The two-controller setups with 16 SSDs have a clear advantage, as this test best shows their ability to accumulate I/O performance. With only eight of the Samsung drives connected to the controller (LSI MegaRAID 9280-24i4e), the performance drops quickly, competing with the other setups only at the lowest queue depths.
In random data transfers, the RAID system is slower than at sequential reads/writes, but it still performs well. The read rate at a high queue depth is between 411.5 and 484.3 MB/s and the write rate between 428.3 and 542.1 MB/s. In contrast to the streaming benchmark, the various test systems perform very evenly in this test. Instead of differing by a factor of two in the read/write rates, the maximum difference here is only about 120 MB/s.
- The 2011 Extreme RAID Project
- The SSDs: 16 x Samsung 470 (256 GB)
- The Controllers: 1 x LSI MegaRAID 9280-24i4e And 2 x LSI MegaRAID 9260-8i
- RAID Creation In Windows
- Benchmark Results: Throughput
- Benchmark Results: I/O Performance
- Benchmark Results: 4 KB Random Reads/Writes
- Conclusion: Second-Generation 6 Gb/s Systems Needed