November 21, 201015 yr Hi, So, I got a couple of new WD EARS 2TB and have been preclearing them with a Silicon-Image 3132 based PCI 1X controller I had from a previous project. I do believe it is PCI-E 1.0 only (Half the transfer rate of 2.0 if I have understood correctly). I observe that when I preclear just the one disk, the read rate in the beginning of the preclear cycle is about 123MB/s. When I start the other one (connected to the same controller) the read rate drops to about 76MB/s on both. I repeated a couple of times to verify. Not totally abysmal performance, but a degradation nonetheless. Am I right to assume this could impact Parity Check throughput as well? Out of curiosity, have other people observed the same with PCI-E 1X controllers? And how about the Supermicro AOC-SASLP-MV8 with 8 disks on PCI-E 4X? I looked over the specs and it does not specify whether it is PCI-E 1.0 or 2.0.
November 22, 201015 yr Hi, And how about the Supermicro AOC-SASLP-MV8 with 8 disks on PCI-E 4X? I looked over the specs and it does not specify whether it is PCI-E 1.0 or 2.0. I have one of these with eight western digital green drives attached (a mix of 1TB, 1.5TB and 2TB) and am getting an overall parity check rate of around 61MB/sec: Nov 18 16:44:27 saturn kernel: md: sync done. time=31970sec rate=61104K/sec Stephen
November 22, 201015 yr I think the slowdown you are seeing is the controller processing or buffering. The PCI 1.0 x1 bus is capable of 250MB/s, but I doubt you will see that actual throughput when you consider all the layers between. I.E. Hardware and Protocol. Also, I have very similar parity check speeds with my setup which consists of 2 PCI-X controllers and onboard drives along with an advanced PCIe RAID0 parity setup. Nov 21 09:42:12 atlas kernel: md: sync done. time=35389sec rate=60681K/sec
November 22, 201015 yr Author As far as I know, the preclear pre-read phase simply reads the disk. Does nothing with the data. And I do have dual-core. So a processing slowdown would surprise me. Or did you mean a processing slowdown on the actual controller processor? Well, I just wanted to check wether this was common knowledghe and expected. For kicks, I think I will try and put one disk on another controller and repeat the excersize.
Archived
This topic is now archived and is closed to further replies.