Results: The Vector 150's Performance Quirks, Continued
Again, starting from a fresh drive, we write sequentially to each gigabyte. As soon as that's done, we begin the write-oriented test a fraction of a second later. That way, all of the addressable capacity is filled, compensating for the behavior some drives demonstrate when reading from a securely erased drive. In a nutshell, many SSDs know that there is no data to read, and just send the data onward. The whole truth is a little more complicated, but in general, you only want to read from sectors that actually contain data, which is why our Iometer testing is performed to a test file.
In this case, we get to see how each drive reads sequentially across its capacity with a 1024 KB read at a queue depth of one, segment by segment.
The SF-2281 processor at the heart of OCZ's Vertex 3.20 again churns out results consistently. Although they're lower than what we might have expected, this is because the data on the drive is incompressible.
As we shift over to the drives armed with Barefoot 3 controllers, we see that performance isn't as high in back as it was up front. Fortunately, the Vector 150 suffers less after the halfway point than the original Vector and Vertex 450. Those two drives get walloped; they're penalized in a similar way with reads as they were in our write testing. It's tempting to think that they're busy doing something else after 50%, but we think the reads are simply coming from slower pages.
We don't have a great hypothesis for the Vector 150's resilience to lower read speeds, but it clearly fares better than the Vector and Vertex 450.
Random Performance Over Time
A saturation test consists of writing to a drive for a specific duration with a defined workload. Technically, this is an enterprise-class write saturation test, where the entire LBA space of the SSD is utilized by a random write at high queue depths.
OCZ is keen to point out that the over-provisioned Vector 150 fares better during prolonged periods of random writes. That over-provisioning helps the Vector 150 shoulder 50 GB of random writes per day, spread out over its five-year warranty. This is largely attributable to a reduction in write amplification, since on-the-fly garbage collection algorithms are more effective with extra space to work in.
We're only showing six hours of writes here, since the Vector 150 enters a steady state quickly thanks to its lower capacity and high performance. I'm also including the original 256 GB Vector, manually over-provisioned to the same level as the Vector 150.
First, we see that the Vector 150 is capable of much higher steady state 4 KB performance at a queue depth of 32 than the Vector. Even over-provisioned similarly, the Vector 150 manages a laudable 21,000 IOPS. The Vector can't match it, since it still has to map the over-provisioned capacity, which incurs additional overhead.
Looking at the breakout showing 20 minutes of writes in single-second slices, there is still moderate variation. But the average still works out to about 21,000 IOPS.
