Seriously, on most current drives (512e) random write performance will be *abso-fucking-lutely atrocious* unless you fix this, and SSDs will have increased wear.
Show this thread
PSA PSA PSA: if you use LVM with USB drives you will get absolutely *abysmal* performance by default (especially on HDDs). You must use `pvcreate --dataalignment 2048s` to fix it. Fix requires re-creating PV and copying all data.
-
-
-
This happens because the USB drive maximum transfer size is 65535 sectors, and LVM in its infinite wisdom decides that's a good alignment value for PVs, thus de-aligning the physical sectors/FS blocks completely, causing read-modify-write cycles for every single small write.
Show this thread -
We're talking 20x performance hit kind of atrocious. 10 minutes for an rsync vs 30 seconds. Seriously. This is fucking bad.
Show this thread -
It seems this may only affect some enclosures (many/most UAS ones?). Check with `pvs -o +pe_start --units s`. If you get an odd or non-multiple-of-8 "1st PE" column then you're affected.
Show this thread
End of conversation
New conversation -
-
-
This Tweet is unavailable.
-
No, this has nothing to do with dd. This is a problem with random writes, not sequential writes. We're talking 20x performance hit on IOPS.
End of conversation
-
-
-
but what if it's in a (properly aligned) partition?
-
The partition is properly aligned. The problem is LVM *de* aligns it. I guess if your partition is misaligned in the *other* direction it might work...
- Show replies
New conversation -
-
-
So a brand new volume/datapool/ect has to be made? That's crazy.
-
All the data is misaligned, the only fix is to copy it all. You could conceivably do an in-place move but I'm not aware of any tool that can do that.
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.