Linux data server - best way to set it up?

feenster99

feenster99

feenster99

Associate
Joined
19 Jun 2003
Posts
1,680
Location
West Yorks, UK
Hi all,
Say you had to build a Linux data/FTP server with Ubuntu 7.04 Server, using an Adaptec SATA RAID card (that supports RAID5, 6 etc), and a bunch of 500GB drives, how would you set it up?

I'm contemplating the best way to do it. To start with, i'll have 5x500GB drives and an Adaptec 3805 card. I was thinking of making one big RAID-6 array (leaving 2 disks for parity), giving me 1.5TB of space. I was wondering then whether to put the OS and data onto this array (instead of having a seperate disk(s) for the OS), and then formatting it with ReiserFS - i've seen it mentioned that this is a happy compromise between speed and safety with Linux.

Are there any obvious flaws in this plan? I was also contemplating putting the array into an LVM group, so in the future i could add another RAID card and 8 more drives and easily expand the space available.

Cheers,
Matt
 
I'd say stick with ext3, I wouldn't say that you'll notice any loss of speed over reiser unless you love benchmarking. It also depends on your file size too, i.e; use XFS for nice big files. As you've said, use an LVM group as it certainly sounds a good idea if you can see yourself running out of space in the future :)

Also, and I may be totally wrong, but isn't Reiser not being maintained anymore? Its dropped from the latest kernel's I believe (at least with SuSE), this is mainly to do with the fact that the author went off and killed his Wife afaik.

Ext4 will be out soon and I have a feeling there is an upgrade path from ext3 to that? (again, I might be totally wrong :D ).
 
Hi,
Thanks for the reply. The filesizes will vary from a few KB, to a few hundred MB each - but there won't be any pattern to this. I think ReiserFS if moving to v4 looking at the website.

Do any of the FS types offer performance advantages when counting files? I will need to keeping track of how big each of the folders are every day, and with a couple of Terrabytes, "du -h *" takes quite a while ;)

Matt
 
feenster99 said:
Do any of the FS types offer performance advantages when counting files? I will need to keeping track of how big each of the folders are every day, and with a couple of Terrabytes, "du -h *" takes quite a while ;)

Matt
Click to expand...


This is the sort of thing you can write a very simple script for and run a cron job to check sizes of folders.

Alternatively depending on the exact setup, you can partition / mount 'folders' at various places and use df -h .

Btw what's difference between du -h * and du -hs?
 
whitecrook said:
This is the sort of thing you can write a very simple script for and run a cron job to check sizes of folders.

Alternatively depending on the exact setup, you can partition / mount 'folders' at various places and use df -h .

Btw what's difference between du -h * and du -hs?
Click to expand...

Just my coding to be honest - did it from memory :S

I do log folder sizes using PHP at the moment (as the script does some other bits, like inserting the values into MySQL and my shell scripting is poor :S), but it can take an hour just to run through 300GB of data and log sizes. I was just wondering if there was a quicker way to do it.

Matt
 
I would use seperate drives for the basic OS install with ext3 filesystem. Then create a RAID5 on the Controller and mount for storage with XFS. This is personally what am doing.

Your / needs to be on ext3 as otherwise I find that the system crashes at startup
 
Hi,
So just put the OS on say an 80GB IDE drive or something? I wanted to put the OS on the RAID5 to keep the hardware requirements down, and for ease really.

There's too many different filesystems under Linux :S

Matt
 
Just have the OS on a small drive, separate to the raid, this cause less problems in the long run. You can make a regular image of the root drive and save it on to the raid, and/or back up on to dvd.

This means in theory should the system disk go down, you can get another disk, re-image and plug it back in. You could use raid 1 but this adds complexity and you said you didn't want that.
 
whitecrook said:
Just have the OS on a small drive, separate to the raid, this cause less problems in the long run
Click to expand...

Hi,
Thanks for the reply. What sort of problems do you envisage could be caused by having the OS on the RAID?

Cheers,
Matt
 
It means your raid is separate to your system. Think of the raid as a device attached to a system. Keeping the 2 apart means that for instance If the raid goes down you can still use the system to fix/repair. There are complications to booting and using the system if certain files are on an inaccessible raid.


If the system disk goes down you still have your raid in tact and if need be you can cleanly transplant to a whole new computer system if required. Just plug in and remount.

Also, if you decide to upgrade, you can upgrade raid array(change disks etc) or the system disks independent of each other (means data is safe if you mess up system disk config), if you have os installed on raid, when you upgrade raid, then you need to transfer/upgrade/the whole system including data to a new system which can be risky.


0.02
 
Like has been said have a couple of small(ish) drives in Raid1 (Mirrored) for the OS/System. Then have your 5 500GB drives in Raid5 and use one as a hot spare. You can then afford to loose 2 of the 500's without causing you too much pain.
 
Yep, i've gone for this setup. I've just used an 80GB SATA for the O/S. I was thinking it could be just as effecient to do a regular ghost of the O/S onto the RAID-5 array every so often. I've never had a good brand SATA disk die on me, and must've used 50+ in the last year in my job. If needed, i can always add another disk in and RAID it.

Just need to work out how to add a hot spare in the Adaptec BIOS, and extend the ext3 volume to take the new discs into account.

Matt
 
You must log in or register to reply here.
Back
Top Bottom