On Wednesday 24 December 2008 20:59:58 Rob OpenSuSE wrote:

2008/12/24 Bob Williams <linux@barrowhillfarm.org.uk>:

...

ATM my /home partition is 1TB on a RAID1 mirrored array. I am going to add a third 1TB disc and convert the array to RAID5, which will give me 2TB usable space.

Is there any performance/safety advantage to breaking this arrangement into several smaller partitions or can I leave it as a single partition?

Performance of what though?

Good question.

For High performance RAID 10, striping 2 mirrors is faster, and more resilient than RAID 5; especially if you will be writing non-sequentially with in place database files. You may run into issues if system partitions were on a 3 disk RAID 5, where with straight mirroring you can always mount the standard device in that situation if needs be, and move back to RAID 1 later. RAID 5 made a lot of sense, when large (for time) reliable disks were very expensive, and cheaper ones arrayed could give better price capacity and performance figures. Presumably you do have another disk used for system stuff.

System stuff will be on its own disk.

When you're just storing and streaming back big multi-media files then, you have not too many problems performance during usage accessing files. Your main risk is very long fsck(8) times on some boots. There can be other advantages to partitioning, it can avoid very long seeks, if files that are accessed simultaneously find themselves in far apart cylinders. They're easier to manage.

I think your point about long fsck times is valid, and one I need to consider. This is a personal home machine, not a data server. Some data in /home/bob is modified frequently, whereas other data is write once, read often. For example, I have about 380GB of music in flac format, which I had thought of putting on its own partition, possibly formatted as ext2.

You can get a lot of advantages of partitioning, without the flexibility disadvantages if you use the LVM. I'd much prefer to get most data out of /home, and keep that for personal settings, and scratch space, rather than 'archive' material, and big downloads etc etc. LVM allows snapshots, which might simplify your back up immensly, if backing up 2TB is every simple.

I dabbled in LVM a while ago, and got into trouble during an upgrade, so I've left it alone. I understand the principle and it certainly makes sense.

Other considerations might be to keep large but rarely accessed stuff, eg) disk backup files from other systems on their own disk, that spends a lot of time spun down, rather than spinning. Utilising more disk arms by spreading files, can also improve performance, without using LVM that tends to result in many disk partitions. A technique I use often is bind mounts to splice in, thinks like web proxy cache onto the right directory in /var which I try to keep smallish size, but with plenty of free space, for fast writes and fragmentation avoidance (helps YaST & rpm).

I have an external SATA device that I can drop 3.5 inch drives into which works very well for rarely accessed stuff.

Partitioning also gives you a chance to deploy filesystems to tasks that they're best suited. XFS is well liked. ext4 performed well in Phoronix tests but is only just about to be become stable.

So many choices. My head's beginning to spin :)

It is indeed safer not to have all your data in one massive pot. Restoring smaller amounts of data in event of a corrupt file system or an accident as root on the disk device, is far easier than one massive filesystem. RAID is not a substitute for backup.

I know that RAID is not a substitute for backup, and run an rsync script every day to deal with that. Is a corrupt file system on one partition of a disk a likely problem, or would other partitions suffer in the same way. IOW are we talking about a disk hardware problem or a higher level software problem?

Downside of multilple partitions, is additional thinking and planning. At the end of the day it has to be your choice, how you organise the data storage, some ppl just want it 'simple' and are unwilling to put any effort in to better organisation.

I don't mind putting a bit of effort into planning a good strategy. Many thanks for your lengthy and thought-provoking reply. And Happy Christmas :) Bob -- Registered Linux User #463880 FSFE Member #1300 GPG-FP: A6C1 457C 6DBA B13E 5524 F703 D12A FB79 926B 994E openSUSE 11.0, Kernel 2.6.25.18-0.2-default, KDE 3.5.10 Intel Celeron 2.53GHz, 2GB DDR RAM, nVidia GeForce 7600GS -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org

Sorry I see I missed a question. 2008/12/24 Bob Williams <linux@barrowhillfarm.org.uk>:

On Wednesday 24 December 2008 20:59:58 Rob OpenSuSE wrote:

...

2008/12/24 Bob Williams <linux@barrowhillfarm.org.uk>:

I know that RAID is not a substitute for backup, and run an rsync script every day to deal with that. Is a corrupt file system on one partition of a disk a likely problem, or would other partitions suffer in the same way. IOW are we talking about a disk hardware problem or a higher level software problem?

The partitions do isolate problems, the driver very much avoids writing outside the range of it's device. We're talking about fairly rare events, but which can be very unpleasant if they happen. Things like managing to write to the disk device directly, overwriting it (a reason not to over use root). There can be software errors. Sometimes memory goes bad. Occasionally corrupt transfers over PCI bus. You may do a bum kernel upgrade. With the RAID you shouldn't have a problem with bad blocks. Just remember RAID 5 in degraded mode, is far more likely to have a failure, than a single disk is. There's also a risk that a failing 'crazy' disk, might corrupt the parity data by reading and writing wrongly, so you do not have a single good copy after all. On con side, it's best to make a backup of your partition table, in case it gets overwritten by another OS, or the disk looses it for some reason, or has it corrupted. A few times I've had to recover data with gpart(8), and it requires patience, with 2TB of RAID5 that would not be my idea of fun Frankly, if the machine got the room for 4 disks, I'd have bought 2 WD 1.0 TB Green Power EADS, and used mirroring especially /boot & /, any crticial data. The disks aren't so expensive and that set up is much less fragile than RAID 5, and faster to recover with a disk failure. When you hit a snag, or need to get your data into another machine most likely with yet higher capacity disks, you can then just transfer 2 disk, not 3 as a complete array. http://www.baarf.com/ though written by DB ppl, has some good general technical explanations on the implications of using RAID 5. And it's characteristics. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org