[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: RAID suggestions?



On Fri, 28 Mar 2008, Stephan Seitz wrote:
> On Thu, Mar 20, 2008 at 10:41:27PM -0300, Henrique de Moraes Holschuh wrote:
>> Then DROP the idea of hw-raid.  Get a damn good SATA/SCSI/SAS HBA, and  
>> use software raid. BTW, damn good means no VIA, SiS, nVidia, or other
>> el-cheap-o half-broken SATA 
>
> Can you give some examples for a good SATA HBA?

No, sorry.  Usually the ones with the latest SIL devices, or those with
hybrid SAS/SATA bridges are good.

> While I???m quite convinced that software raid is more flexible than  
> hardware raid (at least for RAID 1), I know that I can do hotplug stuff  
> with my 3ware (or the PERC 5/i in our Dell servers). And the last time I 

A 3ware board is probably a damn good SATA HBA when in JBOD mode...

> checked with the kernel SATA support, hotplugging disks was not very  
> well supported.

Hmm? It works perfectly, it just complains a damn big lot if you hot-remove
a disk *without* issuing a command to detach it from the logical SCSI bus
first.

What is damn bad is that any late interrupts from the SATA HBA, regardless
of the reason, may cause the kernel to kill an IRQ line, and send the entire
system into a spiral of ugly death.  This is a general Linux issue re.
interrupts, though.  Maybe MSI-capable HBAs avoid this Linux shortcoming...

Note that *any* PCI board using normal PCI IRQs are affected, this includes
any HW RAID card.  Only, HW RAID cards have something else between the SATA
bridges and the host, which will usually eat up stray interrupts :-)

> With my 3ware controller I can use tw_cli or the GUI to rescan for a new  
> disk or to remove it and I use this feature for backup. How would I do  
> this with a ???normal??? SATA controller?

Using the Linux SCSI layer, and mdadm.  Look for the "scsiadd" and "mdadm"
manpages, and also read the documentation on SCSI sysfs (which can do what
scsiadd does using IOCTLs).  Udev can be used for hotplug notification
(insertion).   The hot-UN-plug is the problem, the system doesn't
differentiate it from a disk gone bad yet, IME, so you have to scsiadd -r
the disk before you pull it out.

-- 
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh


Reply to: