I'm setting up our old ReadyNAS 314s for a staff member and have been testing out RAID 5 as the best option for them. So far the results are pretty promising: At least 85MB/sec over ethernet (and that was a driver limitation on my computer, so I'm expecting at least near saturation of the 1Gbps link once I retest), and rebuild/resync times for old 6TB WD Red drives of 16 hours.
The one thing I can't easily test is whether the rebuild times will be significantly slower once the RAID contains large amounts of data (since it would take a very long time to fill up the drives). I was wondering if there is a higher overhead penalty for resyncing data compared to syncing empty space, or if they are effectively the same thing.
I was wondering if there is a higher overhead penalty for resyncing data compared to syncing empty space, or if they are effectively the same thing.
It doesn't matter if the file system is empty or full, the resync takes the same time. The RAID array itself creates a virtual disk (volume) out of the physical disks. It isn't aware of the file system installed on that volume.
FWIW, there isn't much point in using Flex-RAID 5 on the RN314. The default XRAID will use single redundancy, which offers exactly the same protection. In fact, when the disks are the same size it will use RAID-5.
Re: Are Flex-RAID 5 Resync Times Dependant on Data?
Cool, thanks for that. I just wasn't sure if there would be extra parity calculation overhead for actual data. 16 hours is a pretty reasonable rebuild time compared to some of the horror stories I'd read about RAID 5, so I'd like to keep it around that mark.
Also, turns out I meant X-RAID after all, whoops. I guess 'Flex' just sounds more advanced. I was deliberately going for the one that would allow for both horizontal and vertical expansion when this person inevitably migrates to a larger ReadyNAS enclosure down the road.
Re: Are Flex-RAID 5 Resync Times Dependant on Data?
The initial sync when creating the RAID is a lot quicker than subsequent resyncs. As it can efficiently set the RAID up. Whereas when you do a resync e.g. to add a disk to an empty bay it's more complicated: existing RAID blocks need to be read; things need to be moved around; possibly an additional RAID layer added etc. The disks have to be synced sector by sector, so all else being equal the larger the capacity of the disks, the longer it will take.
Whilst the RAID is at a lower level than the filesystem, reading/writing data does use up resources, reducing the resources available for the resync and thus slowing it down.
Re: Are Flex-RAID 5 Resync Times Dependant on Data?
Right. That's definitely more like I was expecting. Of course, I'm not so concerned with the expansion times as I am with rebuild times (in the event of a drive failure and replacement), since expansions are planned and (presumably) remain redundant.
So I suppose my main question is: Would rebuilding the replacement of a failed drive that was filled to capacity see a significant time penalty over initial synchronisation? It imagine it would involve parity processing, but presumably only the new drive would need to be written to and the other drives only read from, as opposed to moving data around. Or am I over simplifying the process?
Re: Are Flex-RAID 5 Resync Times Dependant on Data?
The RAID is at a lower level than the filesystem so it doesn’t matter how full it is. It only matters if you are reading/writing data (e.g. over SMB) as that’ll slow the performance of the resync.
It’s less efficient than the initial sync (have to read from all the other disks and write primarily to the replacement disk) so it will be slower than the initial sync, and can be expected to be significantly slower.
