[geeks] Nifty NAS board (Atom-based FlexATX from Supermicro)

[Nadine Miller]

On May 25, 2009, at 5:55 PM, Sandwich Maker wrote:

> " From: Nadine Miller <velociraptor at gmail.com>
> "
> "
> " I have two Gigabyte motherboards (EP45-UD3P, and G31M-ES2L), both  
> with
> " Realtek NICs (same chip, 8111C).  Both Solaris 10 (though one is u6
> " patched up to May 15th, the other u7, patched up to May 15th).   
> Md5sum
> " on the NIC driver is the *same*, so it's not a regression unless the
> " bug is someplace else in the TCP stack.  One is a trooper, holds up
> " fine under heavy network load.  The other falls off the network
> " someplace between 40 and 60GB continuous traffic via ZFS send/recv  
> and
> " can only be brought back with unplumb/plumb (performs poorly
> " thereafter) or via reboot (rinse, repeat).  It's not the hardware,
> " because the same box held up for a 298GB transfer via rsync using a
> " Linux live CD.
>
> random thought: could be a difference in the zfs tools or drivers.   
> are
> the boards similar enough that you could swap hard drives, and see if
> the failure tracks the os or the hw?

Not sure an HD swap would work.  I'll have to re-review the mobo  
specs.  Iirc their HD controllers are different.  I'm leaning towards  
an interaction b/w driver and tcp stack as there is an outstanding bug  
in Nevada for this RealTek chipset.  I haven't tested with a traffic  
generator (vs. zfs commands) to see if I could get the G31M to fall  
off the network, but that would be a good test to eliminate a zfs  
bug.  Or I could just try copying data using rsync or scp or whatever.

Would that I had support so I could report it as a bug in u7...might  
try to do that under work's contract, though we have no x86 Solaris  
there.  Worth a shot.

=Nadine=