| 1 |
Mike Williams wrote: |
| 2 |
> Yesterday an IBM ServeRAID decided to mark it's 3 SCSI disks as defunct when |
| 3 |
> they are all in fact perfectly fine, giving me a 4am finish this morning |
| 4 |
> after the major hassle of rebuilding, so I'm now heavily biased against |
| 5 |
> hardware RAID, when I know software RAID is fully capable. |
| 6 |
> Plus, mdadm can give you all the information you could ever need, and bugs get |
| 7 |
> squashed quickly. http://bugzilla.kernel.org/show_bug.cgi?id=5181 |
| 8 |
> |
| 9 |
> I think the general consensus is that now CPUs are so cheap, and so powerful, |
| 10 |
> that they can quite easily offset the extra horsepower needed, unless your |
| 11 |
> workload is heavily CPU bound. |
| 12 |
> |
| 13 |
> None of the workloads on any of my servers are heavily CPU bound, so apart |
| 14 |
> from this one server that came with the card (though an acquision of another |
| 15 |
> company), all my RAID needs (on some 16 servers) are done in software. |
| 16 |
> |
| 17 |
|
| 18 |
Both software and hardware RAIDs can and will flake at some point so |
| 19 |
it's a toss up there. I find hardware a bit easier to work with as I |
| 20 |
never need to mess with grub and whatnot to get things to boot correctly. |
| 21 |
|
| 22 |
CPU is just part of the equation in RAID. Assuming I/O is your biggest |
| 23 |
problem having a nice 256MB cache on the raid card can change expensive |
| 24 |
short writes into nice long writes can really help an underperforming |
| 25 |
server. |
| 26 |
|
| 27 |
I'd say if you want raid for better fault tolerance stay with software |
| 28 |
raid. If you also need performance spend the money and get a decent RAID |
| 29 |
card. Do not get the lame ass winmodem raid cards. You'll have driver |
| 30 |
issues and they basically emulate a software raid badly. |
| 31 |
|
| 32 |
kashani |
| 33 |
-- |
| 34 |
gentoo-user@g.o mailing list |
Archives