Testbed and Methodology
In fact the major difference from the hard disk drive testing methodology that we have been using all last year is the testbed upgrade. Our good old testbed on a Pentium 4 processor and ICH7 chipset looked completely out of place in 2011, to put it mildly. The “decrepitude” of our testbed stood out particularly badly in synthetic sequential read/write benchmarks, where CPU utilization hit 100% and the drives couldn’t reach the declared IOps. Even in WinRAR tests we saw that the performance of different drives stumbled upon the CPU speed.
Yes, the testbed was definitely due for an upgrade. But with what? We didn’t want to use an external SATA-600 controller, because this solution wouldn’t ever become mainstream. Therefore, the results obtained on a platform like that would be interesting only to a small user group.
A mainstream chipset from a leading maker could be exactly what we were looking for. AMD fans will have to forgive me here, but I am talking about a different company. It was the P67/H67 chipset from Intel that seemed like an ideal platform for the new-generation testbeds. What did we need from a mainboard? Ideally all we needed were the two PCI-E x16 slots (although we could even do with one if there were integrated graphics) and SATA-600 and USB 3.0 controller integrated into the chipset.
The free PCI-E slot would be used for external controller tests and, of course, for testing SSD drives with PCI-E interface.
As for the CPU, we didn’t need a super-powerful multi-core one: it would heat up a lot and would require an expensive cooler, if we cared for acceptable acoustics. We decided to get closer to the ordinary users and get an i3 model but with the maximum clock frequency. And no Turbo Boost!
As for the RAM, let’s use as much as necessary for the Intel NASPT and Futuremark PCMark vantage benchmarks – 1 GB. If we use more memory, then our Windows 7 operating system will get overexcited and will allocate all free memory for files caching, which will seriously affect the results of those tests that work with files (NASPT, FC-Test).
So, as a result we put together the following test platform:
- Intel DH67CLB3 mainboard;
- Intel Core i3-2120 CPU (3.3 GHz);
- 1 GB Samsung DDR3 PC10600 SDRAM;
- Intel SSDSA2MH080G1GN system drive (80 GB);
- Microsoft Windows7 Ultimate 32 bit OS.
The software part of our testing methodology has also been updated accordingly. We continue to use IOMeter, FC-Test and PCMark Vantage, but we also added Intel NASPT and later on will add Futuremark PCMark07 (unfortunately, we didn’t include it today).
In Intel IOMeter patters we reduced the increment for the Random Read/Write tests in order to better analyze the drives’ performance with data chunks of smaller size.
Defragmentation tests are now performed using Windows 7 built-in defrag-utility. Of course, we know that defragmentation is useless sand even harmful for SSD drives, because it wastes the drive’s writes. But we are extremely curious to see how the new SSDs cope with this type of load.
Now let’s say a few words about the succession of tests. Our primary goal is not to confirm the official maximum performance numbers declared by the manufacturer, but to obtain the performance results closest to what the ordinary user will see on his system, we run all tests as a single block without straightening out the drive between the tests. No linearization, zero fills or Secure Erase. The only thing allowed was a 10-minute break between the tests.
When we took the drive out of the box, we created a 64 GB partition, run PCMark vantage and NASPT tests. Then we delete the partition and create two 64 GB partitions for the FC-Test. Then these partitions are deleted again and we save a previously created image of a 32 GB partition with installed OS, applications and a set of test files onto the drive. After system restart we launch a test script that defragments the partition on the tested drive, restarts the system, compresses the test files using WinRAR tool, restarts the system and extracts the compressed files into a new folder. Next the partition is deleted from the drive and we start a set of IOMeter tests. The test session is rounded up by power consumption measurements.
So, let’s discuss the obtained results.