Re: Web server Partitions

[Alvin Oga <aoga@ns.Linux-Consulting.com>]

hi ya andrew

On Tue, 16 Dec 2003, Andrew Malcolmson wrote:

> On Tue, 16 Dec 2003 17:12:08 +1300, Edward Murrell wrote:
> 
> > The first thing I'd look at doing is moving the default webpage to a
> [ Edward's advice on partitioning the web server}
> 
> I also have a server with 6 SCSI drives and a hardware RAID controller. It
> will be a web server initially but eventually will be a light-load
> database and shell server also.  Should all the paritions be included
> into one RAID volume or is there any reason to put some partitions
> (/tmp? /?) on a non-RAID'ed drive?

that depends on your ability to re-assemble a broken/dead raid system
and still make it work after you fiddled with the broken raid

the answer also depends on what is the purpose of the raid system
	- to protect against 1 disk failure or to increase disk
	read/write thruputs

raid can break due to:
	- (1) disk failures
	- the silly system takes forever ( dayz ) to resync itself
	- too many disks failures renders the entire raid useless
	- data and system on same raid 
		- something goes bad, you lose both system and data
	- whether you can boot off raid5 is a trick question
	  of how you built the kernel and a customized initrd
		( no different than booting off scsi disks,
		( you cant read the scsi disks till you boot a
		( scsi-capable kernel -> fix initrd to solve the problem
	- users will do anything and everything to break the system
	  in the name of convenience and better/faster/easier for "them"

ways around it
	- system should be on raid-mirroring and data on raid5

	system should be mirrored and than stripped  ( better/easier )
	or alternatively stripped and than mirrored

	or the system can be on a non-raided disk and raid5 for data only
		- have an 2nd system disk for backup and go live by
		simply changing its ip# and hostname
		( even simpler )

	- i prefer to keep "system" separate from "user data" (diff disks)

there is no point to raiding /tmp ...
	- if the system dies ... all temp data in /tmp wont matter

	- swap is already "semi-raided" by the kernel
	and if it dies... swap data is generally useless anyway

c ya
alvin