Dumb filesystem idea

Indigo indigo at voda.cz
Mon Feb 5 08:23:24 UTC 2007

> On 2/4/07, Indigo <indigo at voda.cz> wrote:
>> Hello Everyone,
>>   Im about to try a disklayout experiment and I wanted to ask everyone  
>> if
>> Im trying things that are pointless or if I should extend the experiment
>> somehow.
>> Hardware:
>>   Highpoint RocketRAID 2320
>>   2xWD Raptor 74GB
>>   5xWD Caviar 320GB
>> Original idea for the setup:
>>   74GB RAID1 (Raptors)
>>    /,/var,/usr
>>   50GB RAID0 (Caviars[10GB from each - maybe less])
>>    swap,/usr/obj,/tmp,[/var/audit]
>>   1TB+ RAID5 (Caviars[the rest])
>>    /home (or just general storage)
>> The goal is to waste as few fast/reliable space as possible on things  
>> that
>> CAN be lost and to generally reorganize the filesystem by file purpose.
> It looks to me like you are wasting system drive channels. That is,
> IDE can only have two drives per channel, SATA can have one drive per
> channel. SCSI is too expensive to waste on 10GB drives. So while you
> might be moving low-use data off of a high-use file system you are
> losing the ability to have a high-capacicy file system.
>> Known issue is that I'll need some script to recreate the RAID0
>> filesystems when they crash.
> Shouldn't be a prolem with gvinum. Except that some applications
> /will/ crash if /tmp dissipears, and you certainly don't want swap to
> dissipear if it's being used either.
>> Am I onto something here? I feel like running in circles - it's dumb to
>> put /var/obj on the RAID1 where it just eats valuable space. But it's  
>> also
>> dumb to put things on a RAID0 where they will crash a running system in
>> the event of disk failure. I know my idea won't work but I wanted to ask
>> if anyone was playing with similar ideas.
> The trick is to balance your performance requirements and your fault
> tolerance with the data usage and system security requirements. The
> fault tolerance of RAID 1 and RAID 5 are nearly the same, each can
> survice exactly one drive failure. In your example above, the Raptor's
> are fast, but depending on what the system is used for you might need
> that speed in /var, swap, or some other mount point, most of the time
> IO on / and /usr is pretty low. On the other hard, RAID 5 is fine for
> a file server, but if you have a database on that volume you might
> want to go with RAID 10.
> So no, your idea isn't dumb, you just didn't give enough information
> to make a meaningful assessment.

Some more detail then:
The HighPoint card is very decent, it can make the whole setup I described
appear as three SCSI disks (da0-2). Ill make one slice on each and then
partition those slices as I described. So FreeBSD will only see 3 SCSI
disks(74GB,50GB,1TB) it shouldn't see the original SATA disks like it does
with on-board controllers.
da0s1a /
da0s1d /var [mail]
da0s1e /usr [homes,databases,htdocs]
da1s1b swap
da1s1d /usr/obj
da1s1e /tmp
da2s1d /usr/store [public storage/fileserver]

Does anyone know how the system will react to da1 failing?


More information about the freebsd-questions mailing list