On Fri, Dec 11, 2020 at 08:28:33AM +0100, pvane...@debian.org wrote:
>
> Working at a vendor supporting FC, for a time in the FC support
> group, I can only agree with Marco.
>
> It's a whole different world. The compatibility matrix is king,
> anything on there is tested to work, anything not 'is not
> supported', meaning we will try to support you, but the main trust
> of attack is going to be 'move to a supported configuration'. Not
> because the vendor doesn't want to support you, but because there
> are too many moving pieces and we need to reduce complexity.
My knowledge of this is quite out of date, but at least at one point,
the contracts from the large systems integrators might require that
the storage and networking cards that they chose for their enterprise
servers had to be "upstream". However, at least some of the contracts
said nothing about the *performance* of the upstream drivers. So some
of the downstream hardware suppliers would supply "upstream" drivers
to satisfy the letter of the contract, but the "upstream" drivers had
***terrible*** performance (I remember one networking card where the
performance of the upstream driver was roughly was roughly *half* of
the performance of the out of tree / binary driver supplied by the
vendor).
I don't know if that's still the case; hopefully it isn't, but that
can be one other reason why enterprise customers might have strong
incentives to rely on an ancient RHEL just so they can use the
vendor-supplied binary device driver....
- Ted
