I have used Soft Raid to build a Raid 0 of three 2TB disks.
Speed Test with Blackmagic is resulting in read and write speeds of about 400MB/s, what is nearly three times the speed of a single disk.
Building a Raid 5 from the same disks is reducing read spead to 270MB/s what I was suggesting, but write speed is going down to only 90MB/s what is slower than a single disk.
Is this normal? What might be the problem? Is this because I use only 3 disks?
Write speed should be faster, yes.
Make sure Write cache is enabled (RAID preferences)
What computer is this?
Its possible there is overhead from being a slower processor, with minimal cores. RAID 5 performance is dependent upon multi core processing and does better with faster cores (such as i7 vs. i5)
Write cache is enabled.
Its an old (2008) Mac Pro with 8 cores at 2.8GHz 24GB Ram
The 3 Discs are in the internal Bays, and the processor seems not to be too stressed...
Are You suggesting that the Processor is not quick enough to get good RAID5 performance?
We can test this next week with an older Mac Pro
Hi again!
Any results?
Best Regards
We ran basic Black Magic tests using:
Mac Pro 4,1
2 x 2.26 Quad Core Xeon
(8 total cores)
6GB RAM
Here is what we got, default Black Magic settings:
RAID 5 - 4 disks
Toshiba 2TB drives
Reads 430MBs
Writes 350MBs
RAID 5 - 3 disks
Toshiba 2TB drives
Reads 360MBs
Write 340MBs
It is a newer generation of Mac Pro (faster Bus and RAM) then my old 3.1 8x2.8
With 3 WD 2TB Black Caviar drives in Raid5 I get around 270MB/s reading and only 80-100MB/s writing.
A single Drive is giving me about 130MB read and write
In Raid0 i get with the same disks up to 400MB/s read and write
Obviously the Processor is not capable of dealing with a Raid5 appropriate...
I no longer have an older 3,1 Mac Pro so I cannot test this. But it should not create write speeds that slow.
Test a RAID 4 volume, same technology, but only parity on a signle disk rather than distributed. If this is a CPU limitation, it would give about the same result.
The good news I guess is if you ever upgrade the machine, even to a Mini, performance on your volume will jump dramatically.
It is a newer generation of Mac Pro (faster Bus and RAM) then my old 3.1 8x2.8
With 3 WD 2TB Black Caviar drives in Raid5 I get around 270MB/s reading and only 80-100MB/s writing.
A single Drive is giving me about 130MB read and write
In Raid0 i get with the same disks up to 400MB/s read and writeObviously the Processor is not capable of dealing with a Raid5 appropriate...
I have to believe that something is wrong. I set up an Apple RAID0 for two WD-RE 4TB drives inside my 4,1 MacPro. The B'Magic readings with a 5GB load were in the mid 300s with slightly faster reads. I added two more drives and had all 4 bays filled, bought SOFTRAID5, and have been getting almost identical results to the Moderator, mid 300s write and low 400s read. Something must be wrong with your RAID, especially since you got decent readings with your RAID0. Your MP 3,1 is not much difference. Most of my clients have vintage Mac Pros, and I service 2 3,1s and 4 4,1s including my own. In fact, with a fast 2+GB VRAM video card, one of those 3,1 beats my 4,1 in PS CC. From what I understand, the improvements in the Nehalem chipsets in 2009 were mostly with memory type and access and more powerful PCIe slots. I am seriously doubtful that your problem is related to processor speed. My client processes make domes and end up with 700MB+ files. The only thing I can think of is that you are using a slow single HD as your boot drive or don't have enough room for your OS cache. I use a SoftRAID 0 FOR MY BOOT DRIVE. I have 2 240GB SSDs striped using only the first 25% of the RAID drives, although I don't think it matters as much with an SSD RAID. Anyway, B'Magic gives me high 500s for writes and high 800s for reads. (BTW, those striped SSDs are on a full length Sonnet eSATA card with room for 2 2.5" drives and two external ports.) I also have one of the older Accelsior 120GB PCIe card drive, but it's read speeds are horrendous, so I switched the boot drive to the SSD RAID.
My Boot Drive is a Striped Raid of two Samsung 850 EVO SSDs connected to the two extra Sata ports on the mainboard.
Blackmagic is rating it with 500MB/s read and write.
I have no other idea what could be the reason if it is not processor or bus speed...
Just out of curiousity, what if you created a RAID 4 volume and test again?
I am not 100% sure, but I think I tried it and it gave me similar results.
My Boot Drive is a Striped Raid of two Samsung 850 EVO SSDs connected to the two extra Sata ports on the mainboard.
Blackmagic is rating it with 500MB/s read and write.
I have no other idea what could be the reason if it is not processor or bus speed...
Well, your SATA bus (SATA II, 3Gg/s) is the same as mine, and if you are getting 500 R/W. I have older, slower Intel SSDs, I get higher on both, but mainly with Reads. BUT, my SSDs are in the Sonnet card which is SATA III (6G). There are a couple of these cards, available at OWC. The Sonnet is expensive ($280), but the SeriTek one is only ($200), but it was not compatible with my old OWC Accelsior drive/card. You could pop those EVOs into the card, and probably get closer to 700W, and 960R. Plus you get 2 external aSATA ports. I also installed one of those in a 3,1 for a client with 2 480GB OWC Electra Pro SSDs. The catch is that you have only one open PCIe slot (other than the video card with the speed and bandwidth to make those fast cards fly. The other 2 slots are only PCI 1.1 with less lanes.
There's also another approach, outlined by OWC for your model only, and involves getting the Apply Raid card (or perhaps another brand). It has both an internal and external port. If the cable is still available, you can connect it to all the internal SATA ports making them SATA III or SAS, so at least double the speed. Look online on OWC's info database of articles.
If you can afford the time and have the backups, you can try rebuilding the RAID5 once again. Have you degragged those disks?
Note: if there is significant amounts of data on a volume, benchmark tests are invalid. I think he stated this was a fresh volume.