Benchmark Results: I/O Performance
It is probably unfortunate for Intel that I start these stories off with our Iometer-based test patterns. As opposed to a PCI Express-based drive like OCZ’s RevoDrive X2 (OCZ’s RevoDrive X2: When A Fast PCIe SSD Isn’t Fast Enough), or even a standard SATA SSD like the Vertex 3, the SSD 510’s I/O throughput starts off mediocre, and doesn’t really improve as queue depth increases.
The RealSSD C300, based on Marvell’s first-gen 6 Gb/s controller, underperforms the SSD 510 at low queue depths, but speeds up at queue depths as high as eight. There, its eight-channel configuration levels out and no longer benefits from increased concurrency.
Our file server pattern is dominated by the Vertex 3 Pro and Vertex 3. Meanwhile, Intel’s new SSD 510 trails OCZ’s old Agility 2 and Vertex 2 (based on the latest 25 nm NAND ICs).
Intel’s X25-M, even after repeated tests, outright fails this test at queue depths above four.
OCZ’s poor Vertex 2 gets absolutely hammered in our Web server workload, which reads a ton of small blocks to mimic the behavior of a Web server sending similarly small files to client machines. This is consistent with what we must come to expect from the Vertex 2 until OCZ rectifies the ECC-related overhead first discussed in The OCZ Vertex 2 Conspiracy: Lost Space, Lost Speed?
And consistent with Intel’s specifications, the SSD 510’s random read performance does indeed trail the X25-M, witnessed here as lower I/O throughput.
These tests are again dominated by OCZ’s Vertex 3s. The Agility 2 takes a perhaps-unexpected third-place finish ahead of our retail-purchased Vertex 2 based on 25 nm flash. Crucial’s RealSSD C300 inserts itself between those two OCZ drives, topping out when queue depth matches the drive’s maximum number of channels. Intel’s SSD 510 ends up second-to-last thanks to larger random transfers and occasional sequential workloads that favor its superior read performance.