snmpd strangeness

[Jeremy Chadwick]

On Wed, Nov 19, 2008 at 02:37:05PM -0500, John Almberg wrote:
>>> This machine has an Intel motherboard and a hardware raid controller.
>>> From what I can tell, there is some Intel software installed on the
>>> machine that makes hardware faults visible to snmp.
>>
>> That would require Net-SNMP to be linked to that software (or library)
>> directly.  Two things can't just "magically talk" to one another.  :-)
>
> As I said, I really have no idea.
>
> Now that I'm reading more deeply in the notes... the monitoring was  
> supposed to be with IPMI. No idea what that is, either, but I thought  
> I'd toss it into the mix.

Ah, IPMI... it's another one of those technologies which is a great
idea, but often horribly implemented.  The most common use is for remote
management (serial-over-IP, or even KVM-over-IP), access to hardware
sensors (fans, temps, voltages), and for some other monitoring-related
things.  It's very useful -- when it works.  :-)

On Intel boards (native Intel IPMI) it might be great.  There's been a
lot of problem reports with Supermicro's IPMI, and most are IPMI card
firmware bugs.

>> I just hope the card is an actual RAID card and not BIOS-level RAID  
>> like
>> Intel MatrixRAID.  If it is MatrixRAID, I highly recommend you back  
>> the
>> entire machine up and reinstall without MatrixRAID, otherwise when you
>> lose a disk or need to rebuild your array, you'll find your array
>> broken/gone, be completely unable to rebuild it, or kernel panics.   
>> Note
>> that all of this stuff works just fine on Linux; the issues listed are
>> with FreeBSD.
>>
>> Generally speaking, we (the open-source world) have gotten to the  
>> point
>> with OS-based software RAID (e.g. Linux LVM, FreeBSD ccd/gvinum/ZFS,
>> OpenSolaris ZFS) where it offers significant advantages over hardware
>> RAID.  There are good reasons to use hardware RAID, but in those
>> scenarios admins should be looking at buying an actual filer, e.g.
>> Network Appliance.  Otherwise, for "simple" systems (even stuff like
>> 2U or 3U boxes with many disks, e.g. a "low-cost filer"), stick with
>> some form of OS-based software RAID if possible.
>>
>
> That's good to know. I was told just the opposite by the guy selling the 
> $650 RAID cards. Who'd have thunk?

Well, hardware RAID has a specific purpose.  I like them for the fact
that they add a layer of abstraction in front of the OS; that is to say,
some of them are bootable even with RAID-5.  FreeBSD's bootloader has a
lot of difficulty booting off of different things, so adding a layer of
abstraction in front is useful.

For example, take into consideration that you can't get kernel panic
dumps (to disk) using gmirror without a bunch of rigmarole.  I forget
which GEOM method it is, but one of them you can't boot off of easily.
gvinum?  geli?  I can't remember.  There's one or two that the
bootstraps don't work with.  Hardware RAID can help solve that.

> The card in the box is a
>
> Intel 18E PCI-Express x8 SAS/SATA2 Hardware ROMB RAID with 128MB Memory 
> Module and 72 Hour Battery Backup Cache
>
> $625 as shown on the packing list, so I hope it's a good one.

Ah, I think it's hardware RAID, and PCIe to boot.  Yes, I would
recommend keeping that!  What does it show up as under FreeBSD?  I'm
curious what driver it uses, and what your disks show up as (daX or adX;
probably daX).

-- 
| Jeremy Chadwick                                jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |