Hyper-V VM Storage Suggestions

Status
Not open for further replies.

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
Yes, that's one potential fix. I just thought it'd be nice to not have to depend on availability of some other tool, or CLI hacking.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
The other problem with the HPA solution is that there's no equivalent to hdparm in the BSD world. ataidle only handles spindown/AAM commands, we need to be able to issue SECURITY and SET MAX commands. You can do this from gparted-live but it would be easier to be able to do it in-FreeNAS. I do enjoy the necessary option flag to use SET MAX ADDRESS though; --yes-i-know-what-i-am-doing

Back on topic. I'm not sure if I'm doing it right, but so far the iozone testing hasn't shown any appreciable difference between an undersized partition or a full-sized one during "general use."

I also set it up as an SLOG device against a pool and saw some interesting results after copying a huge volume of data. Excuse the mangled graph, but it wouldn't all sit on one.

ExKL7GD.png


After a long, steady run at ~40MB/s it takes a hard nosedive. I'm guessing the sustained writes were too much for it, and the writes managed to outpace TRIM. I cut it off after observing that for a bit.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
That was what happened with the partition covering the entire drive?
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110

Steven Sedory

Explorer
Joined
Apr 7, 2014
Messages
96
Sorry for the delay. So I really can do this at any point now...but the last piece of the puzzle I'm missing is understanding underprovisioning the SLOG. I read through all your thoughts on it. Sounds like it's the right thing to do. With the volume I have setup, what would be the right size to underprovision it to? Is this something that I do when I'm adding the SLOG to the volume, or something that has to be done beforehand?
 

depasseg

FreeNAS Replicant
Joined
Sep 16, 2014
Messages
2,874
It has to be done beforehand. And the size depends on your estimated max pool write speed times 5 seconds. My rule of thumb for a 10GbE link is 8GB. I made mine 16 just to be safe.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,110
Honestly, through my testing I haven't found underprovisioning to make much difference in the case of drives with TRIM. Still trying to verify the one potential case where you consume all available pages (write more than the size of the drive) at max sustained rate, which means that the TRIM commands never have a chance to get in there and clean things up; I suspect that performance there may differ significantly by drive controller.

And the size depends on your estimated max pool write speed times 5 seconds.

IIRC, "max write speed x10" is actually better because your SLOG should be able to hold a minimum of two full transaction groups.

If there's been any tuning done against vfs.zfs.write_limit_shift or vfs.zfs.txg.timeout this will also impact it.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
The reality is that for the typical SLOG device these days being > 100GBytes, even a 40Gbps filer could only write about 5GBytes/sec, so simply underprovisioning to ~60GBytes would be sufficient. Cheat where you can ;-)
 

Steven Sedory

Explorer
Joined
Apr 7, 2014
Messages
96
So I'm finally doing this right now.

I just added the two Intel 750's mirrored as the SLOG, but could not underprovision it first :/ Basically, I couldn't get gpart to see the devices, which is apparently an issue others are having (linux I know, but see https://bugzilla.gnome.org/show_bug.cgi?id=755022). If anyone can show me how I can resize the SLOG now that it's part of the pool (because NOW gpart can see it), or if there is another way, please let me know.

I also added the 3rd Intel 750 a L2ARC.

No problems! thank God.

I will post again sometime soon regarding performance. But again, any info on tricking the Intel NVM's into being underprovisioned would be much appreciated.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,681
It isn't likely to kill you either way, so you could probably just leave well enough alone.

Actually, a more useful thing to do might be to drop one of the 750's out of the pool, wait a year, and then add it back in. As it stands, you've created a SLOG mirror that contains two devices that are likely to fail around the same time.
 
Status
Not open for further replies.
Top