Home

Search Posts:

Archives

Login

January 2014

S M T W H F S
1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29 30 31

"Advanced format" drives are a scourge. The fact that they use 4k sectors is good - but the fact that they lie and emulate 512b sectors is bad.

The issue? When an OS doesn't know the actual sector size, your disk tools end up using the 512b "virtual" sectors reported by the drive that can actually straddle multiple 4k physical sectors. This can cause substantial performance degradation, since to get the the 512b "virtual" sector you need, the drive may have to read two 4k sectors to create an emulated 512b sector. Yuck-o.

The emulation isn't quite as big a deal if your 512b sectors are aligned to evenly match the 4k sector boundaries. It's still suboptimal but it's not a disaster. This is what, e.g., Windows does - and this is why this strategy was apparently chosen by the storage vendors (old versions of Windows/NTFS can't understand 4k sector sizes, so the 512b sector emulation is there to allow these things to work at all).

But Windows' problems suddenly become everybody else's problems, since the storage vendors are now lying about disk geometry. If your filesystem tools are designed with the assumption that disks are telling the truth, they'll use the wrong sector size. And that's where things get sticky.

There are two ways to deal with this in ZFS. The first, and probably easiest, is to download a patched zpool which defaults to ashift = 12 and use that whenever you need to create a pool on such drives. Once the filesystem is created, you can go back to using the normal zpool command instead.

The "right" way to do this is to use the new facility to override sector size via /kernel/drv/sd.conf. The official Illumos wiki has information about this method now as well, but it's a little confusing to piece together how you actually make it all work. Specifically, finding the strings you need to put in sd.conf is really unclear; luckily a user added a comment on the issue tracker which provides a command that gives you just that.

So here's what I did with my Western Digital 3TB "Red" drives, AKA "WD30EFRX":

1) use the command above to get the "magic strings":

$ echo "::walk sd_state | ::grep '.!=0' | ::print struct sd_lun un_sd | ::print struct scsi_device sd_inq | ::print struct scsi_inquiry inq_vid inq_pid" | mdb -k
inq_vid = [ "ATA " ]
inq_pid = [ "WDC WD30EFRX-68A" ]

(If you have multiple disks, the output of "format" and "inq" should point you in the right direction)

2) configure sd.conf with the vendor string and a substring from the product string:

sd-config-list="WDC WD30EFRX","physical-block-size:4096",
"ATA WDC WD30EFRX","physical-block-size:4096",
"ATA-WDC WD30EFRX","physical-block-size:4096";

(Note - I think the "ATA WDC WD30EFRX" is the "right" entry there, but I added the others just to be safe)

3) do a full reboot (reboot -p). You can avoid this by doing the stuff mentioned on the wiki, supposedly, but that didn't seem to work for me. I think I needed to detach and re-attach the disk, but I didn't feel like messing with it and rebooting worked.

4) create the pool like normal

5) ensure that your ashift is 12

zdb -C vault | grep ashift

(use your pool name instead of vault)

I'll say this: the right way seemed a lot trickier in this case. I guess the notion is that in the future sd.conf will be maintained with a list of most such drives so users won't have to worry about it, but for now this isn't nearly as friendly as a simple command line option.

Oh well - gets the job done!

New Comment

Author (required)

Email (required)

Url

Spam validation (required)
Enter the sum of 7 and 6:

Body (required)

Comments |Back