Hi,

This is a long post because I've done a lot of testing and research already. I'm truly stuck! I've got a reasonable knowledge of unix systems (I work as a Solaris 10 admin during the day), although I'm a novice when it comes to compiling stuff and I've never compiled drivers and had little experience with kernels.

I'm building myself up a mythtv server, trying to keep costs to a minimum. I've got everything together and bought myself 3 250GB WD Sata-2 drives which I want to set up as RAID 5. As I've already learned I cannot use the hardware assisted raid mode of the Promise controller, so I've been trying to set up software raid.

My hardware is as follows:
Intel L440GX+ server board, dual PIII 600E processors
1GB PC133 ECC RAM
Ultra 160 73GB disk for /, /usr, swap and /home.
Promise S150 SX4 sata hardware assisted raid controller
3 x WD250JS 250GB sata 2 drives, jumpered to sata 150 mode for compatability
2 x Intel PRO 100S desktop adapters

I've installed Fedora 5 (selinux disabled for ease of configuration). Since I was running into trouble with RAID during the installation, I installed just using EXT3 partitions on the SCSI disk and proceeded to work on the RAID problems later. I've run yum update, so everything should be up to date. Also I noticed the fedora inbuilt SX4 sata driver being loaded during the installation.

So here's where I'm running into trouble. I started with 1 partition on each disk, and using mdadm to create a raid 5 array. Seemed OK - it took a while to build as you'd expect. I set up jfs with mkfs.jfs /dev/md0 and then mounted the filesystem. Cool. Then I started trying to dump my media across to it and the filesystem generated an error message and went to read-only. So I formatted it with EXT3 and tried again, and again errors and read-only.

So... I decided to play with much smaller parititions so the build was quicker and easier. I created a 10GB partition on each disk and created my array. Again, with JFS I got this error (I captured the error this time around)

Sep 26 20:58:24 server kernel: ERROR: (device md0): dbAllocNext:
Corrupt dmap page
Sep 26 20:59:07 server last message repeated 2 times

not much on google on that one!

So I reformatted it with EXT3 and tried again.

Sep 26 21:09:07 server kernel: EXT3-fs error (device md0):
ext3_new_block: Allocating block in system zone - blocks from 3768320, length 1 Sep 26 21:09:07 server kernel: Aborting journal on device md0.
Sep 26 21:09:07 server kernel: ext3_abort called.
Sep 26 21:09:07 server kernel: EXT3-fs error (device md0):
ext3_journal_start_sb: Detected aborted journal Sep 26 21:09:07 server kernel: Remounting filesystem read-only Sep 26 21:09:07 server kernel: EXT3-fs error (device md0):
ext3_free_blocks: Freeing blocks in system zones - Block = 3768320, count = 1 Sep 26 21:09:07 server kernel: EXT3-fs error (device md0) in
ext3_free_blocks_sb: Journal has aborted Sep 26 21:09:10 server kernel: __journal_remove_journal_head: freeing b_committed_data Sep 26 21:09:10 server kernel: __journal_remove_journal_head: freeing b_frozen_data


I got quite a few hits on google, but no resolution that worked for me. I did extensive testing with different raid configurations, including raid 0 on two disks. I changed various bios settings like PCI latency, disabled unused IRQs (serial and parallel ports), changed the sata boards cache to write-thru, but none of these changes helped. I've since undone the bios and cache changes (other than the serial and parallel ports, since I'll never need them anyway)

Anyway, not convinced that all of my hardware was working, I blew away /dev/md0 and the 10GB partititions, and created 3x250GB JFS partitions and mounted them. I dumped over 100GB of data across these overnight with no problems whatsoever. I tested all 3 disks.

So... help please! Has anyone got a S150 SX4 working with software raid on fedora 4/5? What did you need to do?

What does 'cat /proc/mdstat' say, that the array is initiated correctly? (eg. the [UUU] sign)

A colleague of mine has Fedora Core 5 running on RAID 5 without problems for over half a year now, so I know it can be done (eg. not buggy software).

His SATA controller is of this type:
00:1f.2 SATA controller: Intel Corporation 82801GR/GH (ICH7 Family) Serial ATA Storage Controllers cc=AHCI (rev 01)


He is running a boot filesystem aside the raid array (for obvious reasons) and is using LVM for ease of storage management. on top of that he is using Ext3 as filesystem, without any problems..

Can you find any other people encountering problems with the controller you describe on Linux?

/proc/mdstat is fine, as is mdadm --detail /dev/md0. I even waited until fully synchronised before running mkfs and using the filesystem. (and that took about 8 hours on 3x250GB!)

I have come across heaps of problems with the SX4 card on earlier releases, when there was no SX4 sata driver in Fedora. But now it seems that they have a driver, but I suspect the driver may just be buggy.

I'm trying to get gentoo on that machine at the moment, because I did find a post from someone who has successfully set up raid on an SX4 with a vanilla gentoo install... i had some trouble with x packages on my last install so I'll try again tonight with a minimal (cli) install just to test raid. At least that will show if the problem is definitely driver related or not. If that fails I'll try ubuntu and finally old faithful, RH-8.

gentoo = painful!

OK first I'll update with the fact that this error occurs even without RAID. I hammered all 3 disks simultaneously last night with FC5 and recreated the filesystem error. The SX4 driver that comes with Fedora 5 is rubbish, and there is nothing else available for the 2.6 kernel that I can find. I couldn't get any other distro's to work either, I had various issues that I won't detail here.

But there is good news... depending on your opinion of Red Hat 9. I've installed Red Hat 9 with the latest (2.6.0.32) version of the FastTrack driver from Promise. It not only detects the card, but supports the hardware assisted RAID. I've now got a nice 500GB /home partition on my file server and it works beautifully.

For my needs I'll stick with that configuration, as opposed to buying a card that is supported under the 2.6 kernels. I'm only building a mythtv backend database and file server, so having the latest kernel etc isn't really my priority.

Gentoo is not to be taken lightly, but of course I can't (and won't) judge your knowledge of linux...

But.. you could also try Fedora Core 1 ?? While in the Legacy project now (if not even obsoleted), it is still Kernel 2.4 based.


BTW.. good that you have a stable system know, though on some other specs you had thought of.

Gentoo is not painful. It is actually easy, but tedious.

Promise controllers are ok in Windows, but forget using them in Linux. I recommend using any 3ware controller or use Highpoint RocketRAID 1520, Highpoint RocketRAID 1540, Highpoint RocketRAID 1640.

I suggest using XFS because it is a lot better than JFS. JFS does not do any alignment or defragging, so it will become slow over time. Video recording needs continous space instead fragmented.

I guess the pain that I was referring to in gentoo was the time it took to install, once I'd waded through the options. And I have a fast internet connection! I guess dual PIIIs aren't particularly fast at compiling stuff.

I'll have a look at xfs, I was just following the recommendations on wilsonet when I chose jfs.

OK it looks like xfs is not supported by the Promise driver. So JFS it is. And, besides, I don't think my filesystem will slow down that much that I'll notice it across a 100MB network.