Table of Contents
01. 2013-NOV-13: First Hardware Testing & The Noctua NH-U9DX 1366
02. 2013-NOV-16: Temporary Ghetto Setup, OS Installed
03. 2014-APR-01: PSU Mounting & LSI Controller Ghetto Test
04. 2014-APR-02: The Disk Racks
05. 2014-APR-08: Chipset Cooling & Adventures in Instability
06. 2014-APR-09: Disk Ventilation
07. 2014-APR-11: Fan Unit for Main Compartment Ventilation
08. 2014-APR-12: Storage Topology & Cabling
09. 2014 APR-26: Storage and Networking Performance
09. 2014-MAY-10: Sound Dampening & Final Pics
Wait, What, and Why?
So, yeah, another build. Another server, to be precise. Why? Well, as nice of a
system ZEUS is, it does have two major shortcomings for its use as a server.
When I originally conceived ZEUS, I did not plan on using ZFS (since it was not
yet production-ready on Linux at that point). The plan was to use ZEUS' HDDs as
single disks, backing up the important stuff. In case of a disk failure, the
loss of non-backed up data would have been acceptable, since it's mostly media
files. As long as there's an index of what was on the disk, that data could
easily be reaquired.
But right before ZEUS was done, I found out that ZFS was production-ready on
Linux, having kept a bit of an eye on it since fall 2012 when I dabbled in
FreeBSD and ZFS for the first time. Using FreeBSD on the server was not an
option though since I was nowhere near proficient enough with it to use it for
something that important, so it had to be Linux (that's why I didn't originally
plan on ZFS).
So, I deployed ZFS on ZEUS, and it's been working very nicely so far. However,
that brought with it two major drawbacks: Firstly, I was now missing 5 TB of
space, since I had been tempted by ZFS to use those for redundancy, even for our
media files. Secondly, and more importantly, ZEUS is not an ECC-memory-capable
system. The reason this might be a problem is that when ZFS verifies the data on
the disks, a corrupted bit in your RAM could cause a discrepancy between the
data in memory and the data on disk, in which case ZFS would "correct" the data
on your disk, therefore corrupting it. This is not exactly optimal IMO. How
severe the consequences of this would be in practice is an ongoing debate in
various ZFS threads I've read. Optimists estimate that it would merely corrupt
the file(s) with the concerned corrupt bit(s), pessimists are afraid it might
corrupt your entire pool.
The main focus of this machine will be:
- room to install more disks over time
- ECC-RAM capable
- not ridiculously expensive
- low-maintenance, high reliability and availability (within reason, it's still
a home and small business server)
The component choices as they stand now:
- M/B: Supermicro X8DT3-LN4F
- RAM: 12 GB ECC DDR3-1333 (Hynix)
- CPUs: 2 x Intel L5630 Quad Cores, 40 W TDP each
- Cooling: 2 x Noctua NH-UD9X 1366 (yes, air cooling!
- Cooling: A few nice server double ball bearing San Ace fans will also
be making an appearance.
- Case: InWin PP689 (will be modded to fit more HDDs than in stock config)
- Other: TBD
Instead of some uber-expensive W/C setup, the main part of actually building
this rig will be in modifying the PP689 for fitting as many HDDs as halfway
reasonable as neatly as possible. I have not yet decided if there will be
painting and/or sleeving and/or a window. A window is unlikely, the rest depends
mostly on how much time I'll have in the next few weeks (this is not a long-term
project, aim is to have it done way before HELIOS).
Also, since costs for this build should not spiral out of control, I will be
trying to reuse as many scrap and spare parts I have laying around as possible.
More pics will follow as parts arrive and the build progresses, for now a shot of the
(click image for full res)
That's all for now, thanks for stopping by, and so long.