I've been playing around with MDADM and RAID 6 lately. I created and deleted the array about a dozen times as I was playing around with different options but when I re-created it tonight, I noticed that things were a little slow. Before, I was able to copy from my internal SSD to the array and averaged about 120MB/s. Tonight, I'm getting about 10MB/s. ????? Could of sworn I did everything the same. I tried the default 512K chunk size then took it down to 64K and same results. Crap speed.

cat /proc/mdstat returns this:

md0 : active raid6 sdh[5] sdg[4] sdf[3] sde[2] sdd[1] sdc[0]
624637952 blocks super 1.2 level 6, 512k chunk, algorithm 2 [6/6] [UUUUUU]
bitmap: 1/2 pages [4KB], 65536KB chunk

unused devices: <none>

Any suggestions?

Some things I noticed that were a little different. When the array was being created tonight, I got an error about how the GPT is not valid but the backup was OK and will be used. I'm going to delete the array again and will wait to see if somebody can hopefully offer up some step-by-step instructions to see if I can get back to square one. (100+MB/s)

Did you create the RAID on raw devices or you partitioned your drives beforehand?

To me, your bitmap looks wonky - 1/2 pages seems small. Mine is currently 0/22 pages, (raid 5, 3 partitions 3 TB each).

Sorry for the delayed response. I created the RAID on raw devices.

Is there an optimal setting for bitmap? Ultimately, this setup will have 6, 4TB drives in RAID 6. Another question, can this bitmap setting be changed on the fly or will I have to delete the array and make that change when it's created?

Not sure how this goes, but usually the bitmap is internal and gets created/allocated during creation. If things were added/modified piecemeal then it may be a bit off. The bitmap itself is used during recovery (recovery only needs to process the flagged blocks) - but from what I can find, if it is too small it forces excess writes to the bitmap, which can translate into slow writes (as the bitmap may get updated multiple times for each write).

The following commands are SUPPOSED to remove the internal bitmap, then allocate it again:
where N is the size of bitmap chunk. You can try leaving out the --bitmap-chunk to see if mdadm allocates based on the storage size. The default is 64MB, but the manpage also says "or larger if necessary to fit the bitmap into the available space".