I need further HDD advice before submitting order.

[Ryan Stark]

On Wed, 11 May 2005 16:19:33 +0100
freebsd.org at donnacha.com wrote:

> Hi again,
> 
> I posted a question here last week, asking for advice on how I should 
> ask my datacenter to divide up the HDDs in my new server.  Thank you
> to everyone who responded.
> 
I might make different choices in partioning a server in datacenter
enviroment vs. my workstation at home. It is likely you will not have
physical access to the machine, and thus you have to have a technician
to service it for failures - at least. I like to know my
datacenter's and more importantly their technician's familarity with
BSD. I also would like to keep is simple for these techs, and practice
KISS whenever possible. 


> 
> Server purpose: Initially just forums, later sundry other Web apps
> i.e. ecommerce, ticket bookings etc.  Will possibly become a heavy-
> duty email server at some stage.
> 
> 2GB RAM
> 
> 80GB HDD IDE:
> / = 1GB
> /usr = 15GB
> /local = 15GB
> Swap = 4GB
> Unallocated = 40GB
> 
> 200GB HDD IDE:
> 
> /tmp = 2GB (is that enough?)
> /home = 28GB
> /var = 100GB (will inclube the forum databases etc)
> Unallocated = 70GB
> 
This storage scheme, or similar would do fine for such a server, imho. 

Since these are IDE disks, I would also consider the fail rate of modern
large IDE disks (especially) in a datacenter. Should one of these drives
fail, you must ask yourself - does your server have onsite backups? How
mission critical is your server's uptime - for ecommerce, etc? If the
disk dies and you have to reinstall the OS, how long is the
datacenter's turnaround on reinstall(+ restore if you have it)? It is
good to plan in advance for the unforseen.

Many BSD servers I have encountered in a webhost/datacenter
enviroment like a disk mirror setup, and update the backup
disk every day/week/* hours or so. When their system goes down after
hours due to disk failure, the datacenter has special instructions
to just swaps the drives, and re-rack the server. Problem solved. Even
if the problem is not hardware related, you could still ask the
datacenter to swap disks in the event of any server issue that takes
more than one hour to diagnose. If the server is not mission critical
and can stand 24 hours of downtime, then this may not be as important a
consideration. 

This disk mirror method is similar to the altroot method also mentioned.
A mirrored disk vs. a mirrored partition differences? Well, it depends
on how mission critical your server is. Depends on how many disks your
server can hold, and how cheap they are from your datacenter. If disks
are cheap, and there is room in the server - I would suggest a mirrored
disk solution for server failure. If you cannot afford it, don't have
room for more disks, or have onsite backups - then altroot will still
be a lifesaver, take up far less space, and cost nothing additional.


> I'll be asking them to put the both disks in dangerously dedicated
> mode, with each on a different IDE bus.
> 
I personally would not do dangerously dedicated mode for disks in a
datacenter enviroment(I do use it at home), again to keep it simple and
stock for the benefit of anyone else working with the server. Just
like all the other FreeBSD boxen in the datacenter. If a tech does not
know in advance your server's disks are in dangerously dedicated mode,
and has to deal with your server for some reason, he might mistakely
blow away your disk in not recognizing the altered disk device layout. A
more likely result would be the tech taking an additional 15-20 minutes
to figure it out. Again, if the datacenter knows FreeBSD and about
dangerously dedicated mode, then this setup method would not be as big
an issue. The real gains of using dangerously dedicated mode does not
outweigh these potential hazards in my experience.


> I decided not to use GPT because, although it sounds great, it seems
> a little complicated for a newbie like me.
If it is complicated or unfamilar for you, then it may also be
complicated or unfamilar for the techs who administer your server.
Unless they know their stuff, I would not expect them to have a good
working knowledge of GEOM or vinum. Due to this, I would be leary of
using either of these setups even if I did feel comfortable with GEOM
and vinum administration and setup. KISS - keep it simple, it's my
mantra.


>  I would be very grateful for any and all advice.
> 
I realize this opinion does not necessarily add any new advice on your
partition scheme, per se. However, I hope it may have brought up some
additional considerations when outfitting new HDD for a server in a
datacenter enviroment.

best of luck!

-- 
Ryan Stark | syah io com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-questions/attachments/20050511/8d83ac54/attachment-0001.bin