The time has finally come. Linux guys, please come help me set up my RAID5 fileserver

[SiM]

Well you might recall my other threads about this (1 & 2)... I finally have got my hands on 4x1TB seagate hard drives... and I need to set it up now.

Planned spec:
Epox 9NPA+ NF4 Ultra Mobo
Opteron 152 CPU
2x512mb DDR400 ram
4x1TB seagate in software RAID 5
7600gs passive
Gigabit PCIe network card
480W Xclio PSU
Ubuntu server

Possible upgrades available (will sell otherwise):
2x1GB ddr400 (overkill?)
4400+ X2 (overkill?)
GTX280 OC just kidding

Issues/concerns/requirements:

1) 32 bit or 64 bit?
2) Initially I need to set up the RAID5 array with 3x1TB drives then I need to dump the data from the other 1TB onto the array. After that I would like to add the other 1TB to the array. Is this possible/safe? Or am I better off leaving the 4th drive separate and having 3 disk RAID5.
3) I need to make a boot partition seperate from the array. DirectHex mentioned this before here.

I'm quite a linux noob so I might need all the basics explained as this is setup.

Thanks for reading/helping!

[directhex]

for a file server, use 64-bit - none of the usual problems (i.e. browser plugins) will appear.

i'm not too experienced with configuring md, so i think peterb or aidjant will be better at suggesting an answer for this point.

the text-mode "alternate" install CD is actually much more powerful when it comes to partition configuration, so i'd be inclined to use that instead of muddling through the GUI. make a partition on a disk which you set to a small disk to use as /boot (ext3 format), and the rest of the space as part of an MD array. hit save, and the MD array itself will then be listed after your physical disks - set it to a single large / filesystem for simplicity.

[SiM]

OK I am grabbing "ubuntu-8.04.1-server-amd64.iso.torrent" now, is that the correct one? I don't see an alternative version for servers...

[Stringent]

Just be prepared to flex those command line fingers. The server version doesn't install a GUI

[peterb]

MD raid systems appear fairly easy to set up - certainly on Fedora - although it isn't something I have done, other than experimenting by RAIDing a couple of USB flash drives just to see how it worked (well!) I took as my guide "Fedora Linux" published by O'Reilly.

If you can get a book on Ubuntu (OReilly publish one I think) that may describe in great detail how to set it up

A good starting point will be here

https://wiki.ubuntu.com/Raid

The man page can be found here... (dry - but contains the bare essentials.

http://linux.die.net/man/8/mdadm

Sorry not to give a definitive "How to" but

a. Its late
b. While I have played with it, and don't see it as difficult to set up - I'm reluctant to give a definitive guide
c. If I give you the pointers and you research it - you'll learn more

One thing to note is that MDraid raids partitions - so on your 4 drives you would create (say two partitions on each one so you could have two RAIDed drives - using one partition from each drive to give two logical RAID arrays. (or just make one - or however many you want. - so you could have a RAID set up for /home and one for /var for example - but they would be on the same 4 physical disks.

Note there is a difference between mdadm - which is purely handled by the kernel, and dmraid which is managed by the bios (if applicable) It is mdadm that you want to use.

[SiM]

Just be prepared to flex those command line fingers. The server version doesn't install a GUI

I'm sure it can't be that bad... I am willing to learn
I will google for some guides now...

[peterb]

You can only set up madm via the CLI - afaik, there is no gui applet to control it.

You also need a non raided partition (of about 100Mb) for /boot (unless you are using RAID 1) as until the kernel has loaded you can't load the madm driver. (The boot process loads an initial ram based kernel which then loads the main main kernel from the RAID system - the initail ram image contains the madm driver) All this is taken care of when you install madm - but you need to be aware of what is going on!

As for server versions - basically they are all tyhe same but the install scripts install different packages (at least that is true of Fedora - which is my main experience) so you can install a basic desktop package and then download any others that you need later - or at install time. Or install a server package and add a desktop.

That is the great thing about Linux - great flexibility - but a bewildering choice when you first start out. Thats why there is no 'best' distro or 'best' gui - some are easier to install than others, some prefer one gui over another.

If you are initailly planning to run it as a file server in a mainly Windows environment, all you need is SAMBA, which will probably be installed with teh desktop machine package. If not, you can install it anyway. Once you are familiar with it, you can look at installing other server packkages for mail, web or FTP)

You might also want to do a basic install first to get the hang of madm - remember it is partition based - set up a small (say 30Gb) OS setup - then play with RAID. When you are comfortable with it - create the 'production' environement and put it into use.

[SiM]

Somehow I missed your post last night.... Thanks for the pointers
I don't expect a full step by step guide, that would be lazy
But I will post here when I get stuck

Step 1) I'm going to burn the disc

[peterb]

Good luck! Its an interesting (and I found enjoyable) learning experience - with the odd frustration on way!

Burning the disc is a good first step

[SiM]

Cheers Does this look correct?

It warned me that I need a root and a swap partition so I put those on the empty 100mb partitions...

[SiM]

Changed to this now... I am guessing it is still wrong

[directhex]

nearly, there, actually. just set the one you've set as "/" to another swap, and hit the "configure software raid" button at the top.

[peterb]

Can't see the pic (blocked by the work firewall) - as a rule of thumb, the swap partition should be equal to the ram (unless you have not much RAM when it should be double) make the swap about 1Gbyte and you won't go far wrong. 100Mbytes is fine for /boot. for / you will need about 10Gbytes - and you might want to consider another partition for /home. It can be useful to have another partition for /var but a lot depends on what server apps you use and how they use the filesystem.

I use Apache with Mysql and php and I configured it so that most of the data is in /var. That makes backups easy - I just back up that. I also have a separate partition for /home (where user directories are stored) again makes life easirer if I ever have to reinstall or upgrade as that partition is left alone.

So to recap

swap partition - about 1Gb (no point in Raiding that)
/boot (boot partition - cannot be raid except raid 1)
/ About 10 to 15Gb
/var (optionalseparate partition) about 5Gb
/home (optional separate partition) - whats left.

Remember that as far as a user is concerned (and in your role as a user) the location of any branches on the file system is irrelevant - it is just part of the filestore. In your role as the system administrator of course you do want to know where the various bits of the filestore are.

As you probaly know, Linux is not drive-centric - so there is no direct equivalent concept of a path like C:\directory\directory

There is (as far as the user is concerned just /home/mydirectory. Only you as admin know (or care) that /home is on a separate drive or partition.

Edit - just noticed Directhex has got in on this - he's the expert!

[SiM]

Thought the swap was a bit small.... Can't I put the root, var, home onto the array?
So
drive 1: 1GB /boot
drive 2: 1GB swap
drive 3: 1GB swap

The rest of the space goes to the array which will have partitions for root, home, var? Which will be configured automatically I guess? Or can these all be on same partition? Or is this not a good idea?

[directhex]

Thought the swap was a bit small.... Can't I put the root, var, home onto the array?
So
drive 1: 1GB /boot
drive 2: 1GB swap
drive 3: 1GB swap

The rest of the space goes to the array which will have partitions for root, home, var?

correct

[badass]

Is there currently stable, reliable open source iSCSI target software for Linux?
Is this reliable?
http://iscsitarget.sourceforge.net/
If you install an iSCSI initiator on your PC's and you use gigabit ethernet you would potentially have ultra fast access to this box and the PC's would believe the LUN you set aside on the iSCSI target software is directly attached.

Rocketdivision do a free one for personal use
http://www.rocketdivision.com/download_starport.html