Things Get Messy! Three Card Testing
Things always look so prim and proper in our photos of the test bench. In reality once testing sets in, things are rarely organized for long. I do have a penchant for going off on tangents and wild testing scenarios from time to time, just for fun and benchmarking purposes. Above is a pic of our testing as we get into testing three controllers simultaneously. A quick click will show the pic up close.
We are using 2 1882X controllers, and 1 1880IX controller in this test.
Some of the SSDs have to come out of the enclosure and we are using fanouts for direct connections for two of the RAID cards. The expander cannot handle multiple controllers accessing the devices simultaneously. We have configured each 1882 with 3 SSDs, and the 1880 with 2 SSDs. Then we create a stripe dynamic volume in the Windows 7 environment.
Again, we will be using very small test sizes to facilitate cache leverage. This will give us some great speeds, but purely just for fun, to see what types of speeds we can reach.
ATTO Disk Benchmark
Here we can see speeds reaching north of 4GB/s, with read and write speeds of 4.25 GB/s in the 512 and 1024 categories.
AS SSD Compression Suite
The compressible data test is showing us speeds of 4.3 GB/s!
Well, 4.3 GB/s would certainly be plenty for most, but I remain unsatisfied. With 3 controllers I feel that the theoretical maximum speeds certainly haven’t been reached yet, so we pushed on with IoMeter. We utilized a 1 GB test file to ensure cache usage, and two workers, with a QD of 10 each. Pure sequential read access at a transfer size of 256 really started to give us some great results!
We did run into problems as we raised the QD to 12 and higher, as the computer began to randomly crash. We did witness a 4.95 GB/s transfer speed, but it is impossible to get a screenshot of the speed, as the video card was crashing each time that we went over the 10 QD mark. We did experiment with several test sizes and QD mixes, but much over 4.7 GB/s and this motherboard simply cannot handle it.
Even experimenting with several different PCIE packet sizes proved fruitless. As the PCIe bus becomes saturated, the video card is pushed out of its required bandwidth, resulting in crashing.
The X-58 that we are using for testing is getting a bit long in the tooth, and certainly wasn’t intended to be running multiple RAID controllers. A future test bench update to an X79 Sandy Bridge Enthusiast system is pending, so with its Patsburg (server) chipset we will soon be able to push the boundaries much further.