[svlug] Hardware for a new server

[Rick Moen]

Quoting Luke S. Crawford (lsc at prgmr.com):

> Hm.  So if you want to spend a bit more money, and a bit more power, 
> I've had very good luck with the xeon E3-1220L   -  it's a dual-core 
> xeon with a 20w TDP, and it supports up to 32GiB unbuffered ECC.

So, fans again.  I'm sure it's good hardware, but I'm trying to see if I
can avoid needing them.  Also, while I certain am not turning up my nose
at enjoying a glorious excess of CPU capacity, for purposes of my needs
over the past decade the only thing limiting about a PIII/650 has been
limited RAM address space.  A modest CPU step-up will be nice, but I
don't need a huge one.

> Without ECC, I never know if a particular problem
> is hardware, 'cosmic rays'  or software.  ECC gives me a lot more
> visibility into errors;   I sleep much better if I know what, exactly,
> caused that crash last week.

Yeah, funny about that, I historically have held the same prejudice in
spades.  When I first deployed the VA Linux 2230 as an urgent
replacement for a VA Research model 500 destroyed by a freak
lightning/wind storm, I expected that my use of ECC and the 2230's
L440GX+ 'Lancewood' motherboard's support for same would clue me
instantly to RAM problems.  Turned out, it didn't:

http://linuxmafia.com/pipermail/conspire/2006-December/002662.html
http://linuxmafia.com/pipermail/conspire/2006-December/002668.html
http://linuxmafia.com/pipermail/conspire/2007-January/002743.html

Anyhow, any time in the future that I even suspect RAM or
motherboard-RAM-support problems, I'm going to make a beeline to
iterative kernel compiles with 'make -j' to some suitably large number,
ECC or no ECC.  And my biggest takeaway lessons are that I now know how to
identify and isolate a RAM problem very efficently using just Linux, and
that ECC just doesn't catch everything anyway.

> The other thing I'd suggest (if the primary goal is to minimize hassle 
> and if you can spend a few more bucks towards that end) is to mirror all 
> the disks.  All the SSD, too.

Yept, that's the position I've ended up at.

> And mount all disks/ssd in hot-swap bays.

{shrug}

For a home server, I'm willing to turn a screwdriver when/if that comes
up.