RAID 0 strange performance Samsung SSD EVO
I have two Samsung SSD's inside my ThunderBay6 and I'm doing some tests with
- 1 Samsung 850 EVO 1 TB
- 1 Samsung 860 EVO 1 TB
Trim Enable in the OS, both Certified.
But the RAID 0 performance is very strange:
Black Magic 1GB: Read 385 MB/s Write 345 MB/s
SoftRAID 5.7 - driver 5.7
MAC OS X 10.13.6
These RW numbers are far from the numbers I have been seeing on the internet.
I get the same RW numbers individually (non RAID) with the disks.
First always use 5GB file size with Black Magic.
If you are using AJA, then use these settings:
Resolution: 5120 by 2700 5K RED
File Size 64GB
Codec Type: 16bit RGBA
These will give you more consistent results.
With RAID 0 performance should scale smoothly.
Here is something to test:
Skip a slot (move one over one slot) and see what happens.
I did skip a slot and now I have Read 784 MB/s Write 706 MB/s using AJA with the specified settings.
Now I'm curious why skipping a slot worked. What's the explanation?
It is a technology issue. A backplane on a typical 4 disk enclosure has has 2 PCI buses and the controller chip allocates bandwidth across the two buses. When your disks are sharing the same PCI slot, they cannot leverage the extra bandwidth of the disks.
With 4 disks, this limitation does not show up, as the bandwidth is spread more evenly.
Thanks for the answer.
Another technical question.
Do you mind to explain why "always use 5GB file size with Black Magic"?
You want to make sure you are hitting the disk as much as possible. Smaller files can fit into RAM, giving distorted results. Even 5GB is small for benchmarking. (but is the largest offered in Black Magic).
With AJA, we recommend 16GB minimum (64GB preferred) to get the most consistent results.
I see.... Now, some things are making sense.
But, just to make, sure: I can set the Resolution to 4090x3112 4K - Full with a 64 GB Test File Size. For testing a disk, this will make any difference from Resolution: 5120 by 2700 5K RED with a 64 GB Test File Size?
Most of the settings don't make much difference.
What we do is try to get a standard set of settings, so users can compare (or we can compare), what they get with what others get.
So we recommend with AJA:
disk cache disabled (default setting)
Then when you say you are getting a certain result, we have lots of data to compare that to.
I doubt if you change to 4K, etc, the difference will be a lot. What we were looking for were the most disk intensive settings for comparing drives.
Thank you very much for all the help.