Messages in this thread | | | Date | Fri, 11 Feb 2000 22:11:55 -0800 (PST) | From | Andre Hedrick <> | Subject | Re: Mirroring Root Disks |
| |
You have made an eligant point; regardless, I do not do what you propose because ATA ...... the kernel does not yet know how to fake the OS in switch to get a boot in a failure.
I will trump the debate with a $2600.00 ATA-Hardware RAID solution I showed at LWE@NYC.NY 2000.
$2600.00 == 320GB (8x40GB ATA Disks) and an 8-channel Host (3ware). This was a RAID 0 stripe.
Imagine that you can remove the disks and shuffle in any order (on the same card) and it finds-reconstructs the RAID?
SCSI can not touch 320GB in disks for $2600.00 USD.
On Fri, 11 Feb 2000, Martin Bene wrote:
> At 00:48 11.02.00 -0800, Andre Hedrick wrote: > > >> Is there a good way to mirror Root disks. > >> > >> Is RAID-1 only options for mirroring under Linux or something else is > avialable. > >> If RAID-1 is the only option, how safe, fault tolerant it is. > >> > >> I fear setting up RAID for root, as any problem with RAID may make > system unbootable. Any > >> comments. suggestions links on Root disks mirroing ???? > > >Do hardware RAID..........ATA or SCSI but that is you only bet. > > Granted - Hardware Raid is probably the most stable and professional > solutin - it's not the only one however: > > You CAN have software raid for every part of your disk, including /. > > A few pointers on software raid: > > * have a look at the howto on http://ostenfeld.dk/~jakob/Software-RAID.HOWTO/ > * You want to use the raid 0.90 code available as patches for 2.0 and 2.2 > kernels, not the code included in the kernel tree > * With the 0.90 raid code, / on raid (any type - 0 / 1 / 4 / 5 work equaly > well) is no problem, the raid arrays get initialized by the kernel on > bootup and can be mounted just like any other device. > * Swap on raid is currently problematical - I think background > resynchronisation of mirror drives conflicts with use as swap; if you > ensure that no resynchronisation takes place while the device is used as > swap, it should work. Swap on raid would only make sense for mirror or > raid5 to gain reliability. > > The only bit of configuration that's still a bit tricky is where to keep > the stuff needed by the bootloader to get a kernel into memory; usually the > stuff stored in /boot for lilo configurations. There are several ways of > dealing with /boot stuff: > > * have a seperate non-mirrored /boot1, /boot2 etc. on each of the disks > containing mirrors of your / filesystem, have lilo write startup info to > each disk by having a seperate lilo.conf for each disk. > > * have /boot on a raid1 mirror and use a patched version of lilo (patch by > redhat available in the redhat 6.1 source stuff). The patched lilo can > handle raid1 mirror devices and will make each of the devices in a mirror > bootable. > > As for stability: The linux raid 1/4/5 code can handle failing disks > without requiring user interaction; you have to be careful with your setup > however - some errors have to be avoided to allow continued operation in > case of a drive failure: > > * Don't put two parts of a raid device on the same IDE bus (master/slave)- > chances are that a failing master drive will take a slave on the same cable > with it, causing two drives to fail at the same time -not good. Besides, it > wont help performance either. > > * If you're using SCSI, be careful which controllers / drivers you use - > some drivers don't handle failing disks well at all and will lock up your > system or slow it to a crawl by stubbornly trying to access the dead disk > and resetting the bus on failure indefinitely if a disk goes down. > > Bye, Martin > > > "you have moved your mouse, please reboot to make this change take effect" > -------------------------------------------------- > Martin Bene vox: +43-316-813824 > simon media fax: +43-316-813824-6 > Andreas-Hofer-Platz 9 e-mail: mb@sime.com > 8010 Graz, Austria > -------------------------------------------------- > finger mb@mail.sime.com for PGP public key >
Andre Hedrick The Linux ATA/IDE guy
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |