Latest interface: 0.3.1
Latest system: 010
heja2009
User

4 posts

Posted on 20 November 2012 @ 13:36
heja
User

3 posts

Posted on 21 November 2012 @ 06:29
Sorry, somehow my 1st post turned up empty, no idea why.
Anyway, 2nd trial:

I want to build a server to replace the current 11TB QNAP NAS:
FreeBSD, zfs with dedup=on if possible, 2GBit/s via link aggregation,
no DB or other CPU intensive services planned
I'm thinking either mirrored pool of 4x 3TB disks (8 disks total as specced below)
or 5x 3TB in RAIDZ (cheaper of course, but also slower with this config?)

Intel Xeon E3-1240V2 3.4GHz
Supermicro X9SCM-iiF (C204)
Fractal Design Arc Midi Tower
2x32GB SSD for redundant boot
8x3TB (Seagate or Hitachi)
32GB ECC DDR3-1333
1xIBM M1015 PCIe 8-port SATA3
Seasonic G-550 (520W)

any comments?

thanks
Herbert
Darknight670
User

18 posts

Posted on 21 November 2012 @ 11:13
Dedupe will kill your rig, kill your mother and rape you sister without an L2ARC
heja2009
User

4 posts

Posted on 21 November 2012 @ 14:10
Darknight670 wrote: Dedupe will kill your rig, kill your mother and rape you sister without an L2ARC
Thank you, I really appreciate your care for the well being of my family.
And in such colorful language.

If the old lady doesn't serve my favourite dish next time, I might be naughty and nevertheless enable dedup on one of the smaller datasets.

I take it otherwise the config is fine, then?

ciao
Herbert
aaront
User

75 posts

Posted on 21 November 2012 @ 17:54
my default setup is raid 10 (mirrored pairs combined), with a intel 320 80gb for slog and some consumer ssd (samsung 830 usually although I'll be moving to the 840 pro for my next one) for the l2arc.
Compression on for pretty much everything except where I put acronis images
no dedupe. ever. it's a mess. UNLESS you are going to clone filesystems as a vmware image, but the problem is you have to put one vm in each one and mount each one seperately since exsi only does NFSv3

You lose a little disk space this way but rebuilds are superfast, so are simultanious reads. I make one big pool with 8 disks and can run a bunch of virtual machines as well as smb shares without any slowdown.

This all runs on a first gen i7 without any problems. Looks pretty similar to what you have and it works great. Where I run into a few problems are smb shares with a ton of files eating all the arc_meta_used so I had to bump that up.

For the boot disks I just used 2 750gb seagates I had laying around. It's good to mirror that but it's rarely accessed so only bother with ssd to save chassis room, use 2 whatever you have laying around. A neat trick is if you have mismatched size drives, install to just the smaller one, then add the larger one to the mirror.
heja
User

3 posts

Posted on 22 November 2012 @ 01:26
thanks @aaront, the mirrored pairs setup seems like a neat idea - didn't think of it before.

also good point about the boot/swap disks. I will just use 2 older 500GB 2.5" disks.

and yeah, dedup is so tempting somehow, but I understand the use-cases are extremely limited.
I currently play around with an old machine and dedup and it is slow as molasses.
heja2009
User

4 posts

Posted on 23 November 2012 @ 12:49
just wanted to say a final thanks to all.
just ordered the server, went for WD RED disks and no SSDs for now.
will try to resist dedup.
Hakker
User

79 posts

Posted on 24 November 2012 @ 04:09
Just keep track of your LLC records. I had to use wdidle3 in order to preserve my greens. I don't know if the REDs have a better Intellipark implementation so keep an eye out for that :)
aaront
User

75 posts

Posted on 18 December 2012 @ 10:48
Hows it working without ssd? If you use nfs shares you NEED an slog ssd. It also lets the disks keep reading while you are writing. ZFS likes to "breathe". Its an odd concept but basically by having an slog to write to, it can then batch write them to the spinning disks. When I watch my gstat screen, during a big file write to an smb share, the spinning disks do nothing, then every few seconds they do a big dump. The slog ssd gets a constant stream in.
Litis1964
User

2 posts

Posted on 9 January 2019 @ 08:18
This sort of realization brings brilliant forms of working terminology based on a variety of information technology. There are reasons that involved to share the nearest form of submission. No matter what happens in the next but if someone focuses on the current state of acknowledgment nobody gonna let him down, so if you keep in mind this my assignment help review amazing fact no one gonna down your ways
Last Page

Valid XHTML 1.1