Tugger the SLUGger!SLUG Mailing List Archives

Re: [SLUG] debian vs FC threads

Only because I was asked.  :)

Dean Hamstead wrote:

this is flame bait : i cant think of anything that is specific to
redhat and its friends. perhaps there are some strange SAN drivers
which only work on redhat. if that is the case, you need to ask yourself
what sort of life is this hardware likely to have? if i update the
kernel will it make the hardware useless? binary drivers and software
suffer from bit-rot horribly in linux.

No, it's not flame bait, it's just ill-informed (so I stand by my comments
earlier on that line).

Nearly all SAN systems have fibre-SCSI attachment -- this goes for all
of the major SAN vendors -- EMC, Hitachi, Fujitsu, etc.  To make the SAN
switch work in failover mode, you need specific hardware -- usually
EMC/Lightpulse style or QLogic chipset (there are third party OEM boards
using these chipsets) dual-fibre SCSI cards.

The business of automatic-failover and detection of reconnects on these
systems is still pretty much a black art, and all of the drivers to do
it are closed source.  The majority of them are only available for Red
Hat in the Linux universe, while the rest are available for either Red
Hat or SuSE.  No other distro choices, sorry.  There are the beginnings
of an open-source driver in the kernel (provided by Red Hat in fact)
but it just doesn't have the features of the closed-source drivers.
e.g. multipath works but failover does not.

So the closed source drivers are available (at a cost) for every version
of RHEL and most recent versions of SuSE, and they are tied to a specific
kernel version and they don't suffer from bit-rot because there are large
companies being paid significant amounts of money to keep them updated.

I doubt that every major SAN hardware vendor is going to go out of business
because their drivers aren't available on Debian, or aren't available to
people who roll their own kernels.  No, you get RHEL, you install that,
the drivers are available for that version of RHEL (RHEL never updates
its kernel, only backports patches, so the drivers remain good over time),
and you use that.  No Debian, no gentoo, no Ubuntu, and no Fedora Core.

And you aren't going to get major data centers pulling out their SAN
storage units and stringing together heaps of USB drives or something
just so they can run without the binary drivers.  There are hundreds
of millions of dollars invested in this stuff, it's good, it's stable,
and it works.  You want to connect 1000 servers up to 500TB of disk
storage, have it work reliably, and have cluster file systems so you
can have large oracle / OCFS / GFS clusters, with SCSI path redundancy
and load balancing?  This is the way it's done, end of story.