Setting A New Benchmark for Enterprise SSDs
Today we announced our RealSSD™ P300 drive for enterprise applications. Like the C300 before it, the P300 is an incredibly fast drive built to take advantage of the bandwidth available via the SATA 6 Gb/s interface. In fact, it outstrips the bandwidth of a typical SATA 3 Gb/s connection in both reads and writes.
But comparing against a drive designed for client applications doesn’t really make sense. After all, the P300 is built with our high-performance, high-endurance, ONFI 2.1 SLC NAND and was designed with features tailored for enterprise applications (ultra-high endurance, high write-cache-disabled performance, and industry leading steady-state performance, to name a few).
To give you a good idea of where this drive comes in against other enterprise SSDs, we enlisted the help of a third-party testing firm, Calypso Systems, an experienced testing company that’s working with standards bodies to define consistent testing and reporting methodology for all SSD manufacturers (something we’re eager to see happen).
Not surprisingly, they used SNIA-compliant test specs when they ran the P300 alongside two competitors in their “RTP 2.0” test platform. The charts below show how the P300 fared against two of the best SATA-based enterprise SSDs on the market: Intel’s® X25-E and the Samsung® SS805. When you see the results, we think you’ll agree that the P300 sets a new bar for SSD performance in enterprise systems.
4 Comments to “Setting A New Benchmark for Enterprise SSDs”
Leave a Reply
I noticed on the 100% RND writes for 4KB block size, the IOPS are 19561, 7556, 1182 for Micron, Samsung, Intel.
But in the write saturation comparison, which specifies IOPS for “continuous RND/4KB writes”, reading the steady states from the graph I get about 20000, 8000, 5000 for Micron, Samsung, Intel.
Both sets of numbers should agree, and they do for Micron and Samsung, but for some reason your data is showing anomalous results for Intel.
Why is that?
Hello, and thank you for your comment.
The difference seen between the two plots is due to the stimulus applied, and its duration.
In the write saturation plot, a single transfer size (4K) is applied over a relatively long period. In the “mixed blocks” plot, however, several different block sizes are applied repeatedly over a similar duration.
You’re right to expect these numbers to agree. We believe some FW tuning may be optimizing the Intel numbers for strict 4K tests, since the 5K IOPs performance does not hold up when broader block sizes are applied. However, we believe that even the 4K numbers may drop if the test is run for a longer duration.
Because of the differences observed in these two tests with the Intel drive, we are running some longer term experiments to determine whether (and if so, when) the single stimulus (4K) performance will decrease to 1200 IOPs.
It is interesting in that in the write saturation test, the Intel drops below 2000 IOPS at about 20 minutes. When I posted before, I was wondering whether the 100% RND Writes comparison was only doing 4K blocks for about 20 minutes, and so it caught the Intel in the “write pit”.
That write saturation comparison makes the P300 look quite impressive. Is there any chance of showing a similar chart that includes the C300? In trying to choose between a C300 and a P300, it would be helpful to know how they differ in write saturation.
One more question. For the Write Saturation Comparison, are the random writes spanning 100% of the SSD capacity (all LBAs), or is it some fraction of total SSD capacity? If some fraction, what is the fraction?