RAID 5 on Redhat

[Mike Lynch_8]

Hi All

I've just powered up my factory configured Dell PE1800. It is running RH ES3 and is configured with a CERC SATA RAID controller - RAID 5.
Now my knowledge of RAID isn't great..but I'm confused by how the partitions have been set up.

Filesystem 1K-blocks Used Available Use% Mounted on
/dev/sda2 766713356 2726304 725040272 1% /
/dev/sda1 101089 25905 69965 28% /boot
none 1027624 0 1027624 0% /dev/shm


I have a boot partitiion, a swap partition and one huge root partition.

Usually when I've configured a machine I've created separate partitions for /, /usr, /var and so on.

Is this approach not applicable when using RAID 5 ?

Would I be wasting my time re-installing everything and creating several partitions ?

Thanks

Mike

[Steven E. Protter]

Shalom Mike,

Your original configuration is quite applicable, under RAID 5. Having a huge root parition gives advantages and disadvantages.

A big root partition obviously lets you use space in whatever folder you want. It however precludes the creation of paritions with different filesystem and can limit flexibility.

I'm old school. I believe in paritions that make sense and here is why.

Print jos usually get stored in /var/spool

If they are sitting in / root fs, and someome bombs the print queue, the whole system can go down.

If however you take the Bill Hassell(hpux guru) approach, your print spooler is in its own partion mounted on /var/spool and if the print spool fills up the system keeps running.

So careful creation of paritions can and probably should be used. It depeneds on the risks your system faces.

SEP

[Uwe Zessin]

The RAID level shouldn't have anything to do with how you organize the partition layout. Just imagine that the RAID-5 set is one large disk -- to the operating system it is.

[Bill Thorsteinson]

You may have to create the partitions at the
raid controller level. However, it appears
the raid array may be appearing as a single
SCSI disk.

However, I would recommend partitions as
others have noted.

Partman may allow you to resize your
partitions. If not backup, then recreate
your partions before restoring.

It appears you may have enough space on the
boot partion to hold /bin /sbin and /etc
which should give you enough functionality
to repartion and restore.

[Stuart Browne]

Yea, the RAID controller is just showing one physical (well, Logical) disk to the OS. This is fine.

Partition it as you like. Here's how I usually do my Linux systems:

/boot - 100MB
/ - 10GB or so
swap - twice-ram - up to 4GB
/var - the rest

The systems I tend to build don't have users so I don't create a separate /home structure. They also only get security updates, so I leave /usr alone. As it's only admin users who access it, I'm not too worried about users installing unwanted stuff.

But /var in a RH system houses all the logs, the mail/print/etc. spools, the web-site, mysql/postgres database files, so it's logical to have it big.

By the same token, you're system goals might be completely different to mine :)

The RAID controller doesn't care about partitions. It's just giving you one, BIG, redundant disk. What you do with the disk is up to you.

Have fun :)

[Andrew Cowan]

I also muse over this question too because whilst I agree with everyone's comments about the OS partitioning, what happens to something like Oracle OFS?

Oracle says in order to maximize DB performance and resilience you should have your database and logs on separate disks, however this won't happen if your partitions exist within the same array. My solution has always been to either create more than one array, or to have a separate disk for something like a redo-log, and possibly paging, and so long as it is regularly backed-up, and even mirrored, this seems to be the best solution.

[Uwe Zessin]

I'm not an Oracle DBA, but I thought that they tell to S.A.M.E today.

S)tripe
A)nd
M)irror
E)verything

[Andrew Cowan]

Absolutely, I was simply using Oracle as an example, this applies to several application types. The point I was trying make is that partitioning a RAID'd volume does not do what you might expect, and can actually decrease both performance and resillience.

I have seen many instances whereby companies have moved all their storage onto a large SAN and performance has nosedived because they did not plan the distribution of their data, and this caused multiple IO bottlenecks.

[Alan_152]

Right now it looks like you only have 1 physical disk installed (sda) with a default autopartition install (everything on sda2 with the boot efi on sda1) If you were using hardware raid, you'd get something like /dev/cciss/c0d0 as your device.