Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

7666@7666@comp.lain.la (2026-10-02 04:33:16)
i have been conducting a massive operation on pomf's storage which is why this is taking so long

- qcow2 runs like ass on large disks. didn't find that out until after running some tests on the datastore.
 - this is due to the L2 table basically being spread out across the entire volume, requiring disk thrashing to just find files, reducing IOPS to about 20 and bandwidth to a few hundred KB/s. sequential reads/writes not affected but that's not the workload so big problem.
 - the fix is to move to a RAW volume but this requires moving the entire 90TB dataset, again
 - I only have about 165TB of usable storage on this server, 90+90 = 180TB, so i can't actually maintain 2 copies of pomf.
 - so i made a new 160TB volume, thin allocated it, made it RAW, rsynced about 30TB over, verified the destination files, then removed the source files, then i'm doing an fstrim pass to free up the 30TB, then i can proceed with the remainder of the copy and then drop the old disk when it is done vs. another delete/fstrim pass.

yeah this shit is HARD man don't ever run a file host

---Replies---

Blurry Moon@sun@shitposter.world (2026-10-02 04:37:34)
@7666 qcow2 slow as SHIT