I love md RAID
Currently reshaping a 3TB (5-drive) RAID6 with 512K chunk into a 4TB (6-drive) RAID6 with 64K chunk.
Having set
echo 750000 > /proc/sys/dev/raid/speed_limit_min echo 750000 > /proc/sys/dev/raid/speed_limit_max echo 8192 > /sys/block/md1/md/stripe_cache_size
I am seeing /proc/mdstat report a reshape speed of ~40000K/sec. Viewing iostat though shows 5 drives reading at ~50MB/s and 6 drives writing at ~40MB/s - probably about as fast as these SATA drives can manage. So in fact the reshape is proceeding at ~160MB/sec of real data. Fantastic!
Perhaps slightly interestingly, I see slightly lower tps (transactions per second) numbers on two of the drives. They happen to be Seagates, while the others are Samsungs and Hitachis. I wonder whether the Seagates accept larger read/write requests than the others, and the kernel is able to amalgamate multiple requests, so we see slightly lower tps.
Neil and everyone else concerned, thank you very much for all your excellent work!
Cheers,
John.