I haven't had any bad experiences myself with LSI RAID cards - mostly
internal to Dell servers, so not handling very large amounts of drives
or data, but still. We do have some plain LSI HBAs handling large
numbers of drives in JBOD mode (FreeBSD ZFS) where they seem to have
been ok so far. The command line management tools LSI provide are
admittedly pretty horrible!

For what it's worth, the OpenBSD guys seem to favor the LSI controllers
and abhor the Adaptec ones. Doesn't necessarily translate to SL
or for other use cases, so I'm certainly not trying to negate Nico's
experience, but I found what they had to say interesting. I haven't
used Adaptec in a long time so I don't know.

What I've found with all RAIDs is the drives themselves can be critical.
We were buying a bunch of RAID storage a few years back where the PIs
wanted to maximise their volume by using 2TB drives, which at the time
were only just becoming available. Had some interesting conversations
with our vendor, who said that they can test whatever drives we want,
but you simply can't tell which will be "good" until 6 mo to a year down
the road - presumably when media errors start to develop and their
response to that (not just TLER). As it happened they and we tested
these early WD RE4-GP enterprise drives for some time with no apparent
issues, but they have proved very problematic over longer term... in
ways that the same arrays populated with different drives have not been.

Most of our external RAID boxes are based on OEMed Areca controllers btw.
They've been pretty good for us.

Graham

On Thu, Nov 15, 2012 at 08:32:56AM -0600, Ken Teh wrote:
> Thanks Nico!
> 
> Anyone else want to comment?  I'd like to hear people's opinions about such
> issues.  When one is not 100% involved in such efforts, it helps to have input
> from people who are or have dealt with similar issues.
> 
> On a related note, this is probably the last disk array I want to put together
> myself. I discovered to my exasperation that I needed enterprise class drives
> because of the TLER effect.  The next disk array I buy will be a ready-made
> appliance.
> 
> 
> On 11/14/2012 08:58 PM, Nico Kadel-Garcia wrote:
> >On Wed, Nov 14, 2012 at 12:37 PM, Stephen John Smoogen <[log in to unmask]> wrote:
> >>On 14 November 2012 10:20, Ken Teh <[log in to unmask]> wrote:
> >>>The common thread is I/O to a MegaRAID raid5 device.  Which is cause for
> >>>concern since the primary function of both machines where I've encountered
> >>>this problem is file-serving.
> >>>
> >>>Perhaps I am just unlucky and have 2 bad MegaRAID cards in a row.  I'm
> >>>trying
> >>>to understand this better, figure out if I am doing something wrong.
> >>
> >>Well there are a couple of issues this could be:
> >>
> >>1) You are asking more than the MegaRaid is meant to do... it may be
> >>running out of cache, or other resources.
> >
> >My experience with MegaRAID has been *horrible*. Poor driver
> >compatibility, awkward and destructive firmware, and deceitful
> >specifications only start the list of horrible failures. Their best
> >technological use is as doorstops.
> >
> >>2) The megaraid is still rebuilding its array beneath and you are
> >>hitting a locking problem because it hasn't finished what it needs to
> >>do before you ask it to do something else (really sort of #1).
> >>
> >>Most of the time you will need to install the proprietary Megaraid
> >>tools to see what is going on under the disks to find out.
> >
> >See above: good luck getting those tools working!
> >
> >Every hour you spend waiting on those things to come to their senses,
> >or trying to debug them, is an hour wasted on problems that may not
> >ever be solved by your efforts. I encourage you to replace them with a
> >better quality manufacturer: Adaptec makes very solid, not too
> >expensive controllers, and Rocketport remains the cream of the crop.
> >

-- 
-------------------------------------------------------------------------
Graham Allan - I.T. Manager
School of Physics and Astronomy - University of Minnesota
-------------------------------------------------------------------------