Re: Bulletproofing a Linux server
- From: Ignoramus24141 <ignoramus24141@xxxxxxxxxxxxxxxxxxxx>
- Date: Mon, 25 Feb 2008 13:31:39 -0600
On 2008-02-25, Chris Cox <ccox_nopenothis@xxxxxxxxxxx> wrote:
On Mon, 2008-02-25 at 11:13 -0600, Ignoramus24141 wrote:
I am building a server and I want to make sure that it is
"bulletproof", which, to me, means that it can last a long time and
survive adversity without too much attention and would have a great
uptime.
It will be hosted in a datacenter, and I do not want to visit it too
often. My current server is 3+ years old, and I saw it exactly once
during this time, when I added extra memory to it.
The server is still around, but its drives are not mirrored and in
fact paired with a big partition that spans them both. (dumb idea) So
I am in a precarious position. Plus it is too slow.
The new server, I want to last 5 years comfortably.
To that end, I have or plan on the following features.
1. Three pairs of mirrored drives managed by a 3ware controller (disk
reliability)
Mirroring is good. 3 separate mirrors... perhaps overkill, but
depends on what you're doing.
I mentioned it in a separate post, I want root and swap on a separate
pair, websites with guaranteed fast response time on another pair, and
"everything else" on the third, slow, pair.
2. Active CPU Cooling (with fans)
3. Use cpufreqd to reduce CPU speed if overheats anyway despite active
cooling
Can be difficult on good hardware... you'll have to look into
this a bit.
The way I see it, assuming good fans, it would allow for fans failing.
4. Use smartmontools to monitor hard drives
Again, this may be difficult, you'll have to do your homework
on this one as well.
If you're plan is to completely ignore the box, I'd get
a RAID subsystems with a massive pool of spare drives. Otherwise,
just the fact that you are using RAID should give you enough
opportunity to fix things up (with minimal monitoring).
That's how I see it also, I will have a few days to order
replacements.
5. Get a little too much on the performance side: 15k RPM drives for
things that matter, and 16 GB of RAM. (RAM being so cheap today, this
is a no brainer, extra 8 GB of registered 667 MHZ memory cost just
$287)
15K drives are going to be less reliable. The larger the RAID
is the less the rotational speed is going to matter. Of course,
now I'd be talking RAID 6, 10, 50, etc. instead of just
RAID 1.
Extra memory is fine.. but perhaps overkill. All depends on
what you are going to be doing. Memory isn't as cheap as you might
think... especially on a server box.
Extra 8 gigs cost me just $277, hard to beat that.
Memory has become so dirt cheap, nowadays, it is mind boggling.
6. Get an IPMI card, so that even if the computer crashes, goes down,
cannot boot etc, I can recover remotely.
Yes... you can do that... but since this is Linux, I prefer to
go the serial route and use a serial console server. Just my opinion.
I have used both styles. I think serial is cheaper and more
flexible.
Has anyone here used IPMI?
My server has a console only interface (no GUI, gdm disabled).
Is there some client tool allowing me to remotely interact with a dead
server? (such as going into BIOS settings, reboots and so on)?
i
.
- References:
- Re: Bulletproofing a Linux server
- From: Chris Cox
- Re: Bulletproofing a Linux server
- Prev by Date: Re: Bulletproofing a Linux server
- Next by Date: install software - how?
- Previous by thread: Re: Bulletproofing a Linux server
- Next by thread: install software - how?
- Index(es):
Relevant Pages
|
