Hey,

Might be asking for impossible here but just wondering if there's a way I could improve cache offloading/mover to array, atm without parity (yeah ik ik) which should remove the biggest obstacle, and I know for a fact that most filled handles this quite well in a properly balanced least free disk across array as it'll bounce between disks happily saturating cache reads, seen it happen and was happy with it.

in a scenario where I've added new disks and balance is off atm and I dont feel like manually fixing it, is it possible to achieve similar to above behavior (read cache to write multiple array disks).

Kr

Mizerka

What you want is not supported.

However if you have no parity drive, then writing to array drives is reasonably fast anyway although limited by individual drive speed, and also you can write multiple drives in parallel without them impacting each other's speed.

In the more usual situation where you do have parity, writing to multiple array data disks simultaneously will not improve performance, and may even degrade it, since parity must be written at the same time each array data disk is written. So could actually cause more seeking on parity than just writing one disk at a time.

This is one reason Most Free is less efficient than Highwater.

hi both, yeah I understand that having parity is the big bottleneck and impact on array with multiple write could cause with parity having to calc each block etc.

and yes writing without parity is reasonably fast, best the single disk can do basically about 180MB for my current most empty disk. with balanced free space across array and with most-free share, its a lot faster, I assume its thanks to split levels rather than writing single files and swapping dest targets, it can write entire subfolders at once.

Thanks, I kind of knew what the answer was but wanted to double check

actually I think I can think of another hack way of doing this but I dont feel like going down that path... with "preseeding" destination folders, by looking up cache directories, creating empty paths to /mnt/disk../ paths, with correct understanding of my current split levels (its somewhat deep atm), it should achieve read one write many scenario.

original problem btw, is that with 2tb cache it just takes a while and cripples new content ingests as cache isn't being used during mover and both cache and new data want to write to the same disk

got util in line with other 2 disks, this is the behavior of cache>array mover with most free shares and without parity;

eventually it will allign with rest of array and bottleneck much closer to cache read rates