TEST BENCH AND PROTOCOL
In testing the Intel DC S3700 SSD, well be using Red Hat Enterprise Linux and/or CentOS, as we do for much of our enterprise testing to date.
Linux has less overhead and is generally more flexible when it comes to evaluating performance. That said, our enterprise test bench is OS agnostic.
Well apply a few new standardized testing techniques in addition to some of our older test protocols.
We want to isolate and explore the individual performance of the review drives as accurately as we can.
As the test bench evolves, we hope the result is a more tangible, relevant performance evaluation.
A special thanks to Asus, Crucial, OCZ, and Fractal Design for sponsoring our Enterprise Test Bench.
To specifically measure latency, we use a series of 512b, 4K, and 8K measurements. At each blocksize, latency is measured for 100% read, 65% read/35% write, and 100% write/0% read mixes.
Intel specifies some of the S3700’s latency characteristics in their datasheet. Typical latency is around 50us for reads and 65us for writes. Not much other detail is given, except that those measurements are performed with sequential transfers. Sure enough, the typical write latency for 4Ks is under that 65us threshold at 49 microseconds. These QD 1 latency measurements are performed after filling the drive twice with 4K writes at QD 32.
Average read latencies are a bit higher, averaging 120us for 4K reads. As is the case for most every drive, 65% read/35% write latency is worse than either the pure read figures or pure write.
Maximum latency measurements at QD 1 are a sort of worst case scenario. These metrics represent the slowest IO among all the transactions. 99.95% of all IOs will complete far closer to the average than the max. Fewer than one out of every thousand IOs will take anywhere near the max numbers above to complete.
At QD 1, Intel says 99.95% of all write IO will take less than 500us to finish. So after a six hour QD 1 test, we recorded the latency statistics to test the claim.
|20.00th Percentile: 31us|
|40.00th Percentile: 32us|
|60.00th Percentile: 33us|
|80.00th Percentile: 39us|
|99.95th Percentile: 498us|
Completion latency is indeed under 500us 99.95% of the time. Better yet, 80% of all write IOs tested over six hours completed in under 40 microseconds.
Awesome product again, but price is very strange its equal to capacity i mean 2x100GB is exactly the same as 1x200GB and so on. For enterprise server more capacity is logical (save power & room), but for mid range servers its a kill – deal to buy smaller drives price/performance it will be a lot more then awesome.
Are you still using IOMeter to generate you test data? Also, are you testing the drive raw, or with a logical file system? Great review!!
On page 5, it’s described that this SSD uses deduplication.. is it right? Because I don’t find Intel has mentioned it elsewhere.. thanks for the great review.