I recently bought 5 external seagate drives on sale which all had the new Exos 2x14 mach 2 drives in them. I was excited as the speeds shown were quite mind blowing for HDD’s.
I put them all into my Synology DS1821+ with SHR-2 raid configuration with BTRFS filesystem.
I connect to my nas through a X520DA2 10gbit link which when testing with iperf3, will easily saturate all 10gbit (had to do some driver tuning on the windows side).
When copying files to and from my Synology, the transfer speed will cap at 300MB/sec-400MB/sec. Now these speeds are not bad per-say, but I was expecting much higher due to the fact they’re dual actuator drives and the amount of them that I have.
Does anyone have any clue why the speeds aren’t at least twice as fast as they are now? Is part of this just growing pains because of the new dual actuator technology in sata drives that Synology hasn’t yet optimized for?
Speeds of your drives are twice that of single actuator HDD’s. Regular HDD’s rarely top 200MB/s, some may get 250 in vendor-specific, highly optimised tests.
Also: SATA has an upper limit of 600MB/s, but due to overhead in communicating 550MB/s is a more realistic limit. That terrain is reserved for SSD’s though, provided their controllers are fast enough.
Yeah I figured it wouldn’t scale absolutely perfectly according to manufacturers claims, but I was still hoping for a bit more. Is the SATA upper limit of 600MB/sec per drive or for all drives connected to the controller?
If it’s the former, shouldn’t I be seeing at least 500MB/sec considering this is 5 drives in what is effectively a RAID 6 array? Excluding 2 drives for parity, that’s still 3 drives x 2 actuators per drive. That alone should saturate my 10gbit link with ease.
Only if you partitioned the drive (SATA) or using both LUNs (SAS). The second actuator is responsible for the second half of the drive. So unless you write or read stuff to/from both ends of the drive, there is no difference to a regular disk.
Once the drive is filled and fragmented, you’ll see both actuators active more often. Or use Wendells partitioning and cut the disk into two partitions. Which can be dangerous because you lose two drives if one HDD dies. So check your RAID config and failure domain.
If you can make a RAID50 out of them, that would certainly improve things for both redundancy and performance. In ZFS speak, that would be 2x 3-wide RAIDZ1 vdevs
I’m pretty sure the problem is that DSM isn’t smart enough to know how to use both actuators on the hdds at once. This is due to how the dual actuator sata drives are presented to the host, as one big drive, the first half of 7TBs of LBAs on the drive are served up by one actuator and the second 7TBs of LBAs are served by the second actuator. Under normal/traditional hdd access, the host is going to just sequentially go through LBAs, which in this case is going to only ever use 1 actuator on the drive unfortunately.
The dual actuator SAS drives are setup in a more sane way, showing as two completely different logical drives, one for each actuator, but the sata drives couldn’t adopt this approach because the sata specification is much more limited.
To take advantage of both actuators on the sata DA drives you’d need to manually partition the drive into two halves.
It’s funny you mention this, because I did run a defragmentation after copying my files onto this newly built nas, and I did notice performance actually went down (but is much more consistent), after defragging my storage volume.
It’s counter intuitive, but does a “refragmentation” concept exist? It seems like this problem might be able to be solved by software that can spread files out evenly across all drives.
As the others have said. To get the double performance from each drive you have to partition them into two 7tb drives each.
5 phy drives would be seen as 10 seperate drives of 7tb (pre-format).
Then tell DSM to stripe the drives so data is spread over all 10 drives.
Hope that makes sense.
Basically these are two in one hdd’s and you need to partition them as such so the OS uses it that way.
Wendell has a decent video here explaining these drives.
What might have happened is that during defragmentation the files were coalesced to an area closer to the center of the platter which is read at a slower speed.