May 14, 2013
David Szczur
SSD in Their Native Habitat

intel_vs_kingston

It is no longer a debate whether or not solid-state drives are ready for prime-time, or if there are any performance benefits you yourself can realize from adding an SSD to your setup. Be it a cheap, light-purpose netbook, a desktop, or a mission-critical quad-processor server, an SSD will absolutely show you some performance gains in one way or another from a traditional mechanical HDD. With recent advancements in MLC NAND flash (Multi-Layer Cell memory used to create most SSD’s), They’re becoming ever more affordable, available, and stable by the generation. And yes. They’re getting much better at what they do best. IOPS all day.

For my testing, and you’re evaluating pleasure, I used the Kingston E100 and Intel S3700, two new SATA III 6gbps 2.5” SSDs recently released to the market. I focused mainly on the 100GB models from each manufacturer because I had several of each. These are both next-gen parts with years of innovation built into them, the E100 able to reach 535MB/s and 500MB/s sequential reads and writes, and up to 83,000 max IOPS, and the S3700 slightly lesser at 500M/s and 200MB/s sequential reads and writes, with up to 75,000 max IOPS. These are, of course, manufacturer specs derived from their own respective internal testing, and not necessarily indicative of real world usage. I was not surprised they performed equally well on their own SATA III port, with the E100 coming out at a ~15% better write performance on average. Those results were great, but not very interesting. I really wanted to see what these drives could do when pushed to their limits.

I chose to install 2 of each model into their own Intel E5620 server chassis with an LSI 9266-4i 6Gbps card and (4) 6gbps ports. Each of the two servers were deployed in a RAID 0 configuration and installed with Windows 2008 Standard x64 for easy results. RAID 0 was chosen, because this stripes the data equally across both SSD’s in the array, essentially treating 2 drives as 1. This is horrible for data security, as one disk failure destroys the entire filesystem. However, for gauging performance of two identical drives, this would truly push their limits. The first test I ran was IOmeter, an open source HDD benchmarking program. I ran the entire suite of tests, alternating randomly from 512 Byte files written to 25% of the disk, to 32KB files written to 100% of the disk, for 8 hours. This benchmark was to show the stability and consistency of the drives under constant load. Toward the beginning of the test, the RAID’ed Kingston E100 drives were hitting 445MB/s write speeds with the Intel S3700’s trailing at around 410MB/s. Performance slowed in the following hours and by the end of the test, the drives were still in close contention.

graph

As you can see from the results, the Kingston E100 gets only slightly higher average throughput per second, and IOPS are almost even as well. An 8 hour burn in proves though the E100 may perform better in small bursts, overall Intel still holds the crown for consistency. It is worth noting also that neither RAID arrays suffered a single error throughout the test. In my experience there is usually at least 1 or 2 errors, telling me these drives are likely quite more stable and better with error correcting than previous generations.

The next test I had chosen was CrystalDiskMark, a freeware program written by Kirino Kasumu for general use that has gained a small degree of fame amongst bench testers for being particularly good at benchmarking SSD’s. This test uses a predetermined capacity of the disk to write and read sequential and random 512KB files, 4KB files, and 4KB files with a Queue depth of 32. Without getting too technical, the queue depth set to a higher number such as 32 is simulating hundreds of users/hosts demanding use of the drives at the same time. The results I got from this benchmark were not as expected.

crystaldisk

Right from the beginning, this testing method favored Intel’s S3700. The Kingston E100 barely manages some higher numbers in sequential and 512KB Reads, scoring a nice 1028 MB/s and 1837 MB/s respectively. It seems though in purely writing random and sequential files to the disk as quickly as possible, the Intel S3700 is the clear winner. Even when introducing substantial queue depth to the operations, the higher factory spec’d Kingston E100 can only barely keep up with the formidable S3700.

Keep in mind, all tests results given for the purpose of this blog post are performed on two identical drives in RAID 0 on a very high performance RAID card on an enterprise server platform. For a typical single drive configuration, the Kingston E100 is a much stronger performer. For a small RAID array built solely for maximizing I/O with no regard for anything but performance and consistency, Intel’s S3700 is a great next step from the 520 series, which in my book currently holds the best price to performance ratio of any SSD on the market.

Leave a Comment