January 3, 20188 yr I have around 3TB/day while copying from reiser to xfs Source/Destination are both HGST 6TB disks (new) CPU is Xeon E3-1270 V6 with 64GB RAM HBA is Adaptec 1000-16i 12Gbps I see in Stats that CPU is mostly below 20%. And Disk utilization overall is 100 MB/s with sporadic peaks up to 200 MB/s
January 3, 20188 yr I thought it should be better. Is it a normal behavior? Do you use normal or reconstruction writes? For normal use (copying from external or cache to array) the latter are faster, but copying between array disks both will be relatively slow. Normal: this will result in read-modify-write cycles for parity drives, so your speed will be limited by those. Reconstruct: in this mode there will be only writes to parity, but now the heads on the source drive will have to travel a lot. On the source drive the head has to go to the next sector to copy for reading source data then it has to go to the equivalent sector of the destination drive in order to read the data that is needed for parity calculation. Of course the drive’s controller will do some caching, but there will still be a lot of head movement involved. I would expect the usually slower “normal” mode to be a bit faster in that scenario due to less head movement, but none will be fast. If you have the statistics tool installed, you will probably see, that there is not a continuous stream of reads/writes, but fluctuations due to the effects described above.
January 3, 20188 yr Author I am running rsync as per this https://lime-technology.com/wiki/File_System_Conversion#Mirroring_procedure_to_convert_drives No reconstruction in progress I see rates as low as 30 MB/s.
January 3, 20188 yr No reconstruction in progress I see rates as low as 30 MB/s. I am not talking about array reconstruction, I am talking about writing in parity reconstruct mode, also called turbo write mode. In this write mode instead of doing read-modify-write on the parity drives, all data drives except the one being written to are read and the parity is recalculated and written directly. Avoiding read-modify-write is usually faster, at the cost of having all drives spun up. But not for copies inside of the array.I did a disk to disk copy in an array recently and while I didn’t check the exact speed I think I landed in the same ballpark.
January 3, 20188 yr Author Thanks, I forgot about this No, I don't want 'turbo mode', otherwise I wouldn't go with unRaid
January 3, 20188 yr Community Expert 47 minutes ago, tstor said: I would expect the usually slower “normal” mode to be a bit faster in that scenario due to less head movement, but none will be fast. In my experience Turbo write is still faster in this scenario, though not by much.
Archived
This topic is now archived and is closed to further replies.