Re: TCP RDMA option to accelerate NFS, CIFS, SCSI, etc.

[Alan Cox <alan@lxorguk.ukuu.org.uk>]

> Given you are concerned about Appletalk and PCI, perhaps we just
> need to agree you are addressing a different performance level of
> storage system.  I see no reason for hardware support at Appletalk

No we are addressing the same level of performance in a general purpose OS
the difference is my case doesn't need weird protocol hacks. When working
with a dedicated storage system RDMA becomes even less interesting because
you code the stack to the needs of the filer. You don't even need an MMU
on such kit

> proposed NIC if this is really to be taken as a serious alternative
> to an RDMA option.  I for one dont find it a competitive design
> (it is also one that was previously considered.)

I don't find RDMA a credible useful solution at the protocol level. It
doesn't offer any visible advantage, it complicates the stack futher thus
punishing the majority in the interest of the few.

I'm interested in why you think such a NIC wouldnt work, Providing you have
interrupt mitigation I see no reason for it to fail. I'd be interested
to know what the flaws in that technique were in your eyes.

> doing this, so I regard this as a done deal.  (I'd hate
> to be a high-speed NIC vendor that cant do this.)

But can you do it at $6 a part in volume ? Thats what the other 99.9% of the
people care about.

> The only issue is how to handle the next level of 
> protocols that have high performance requirements, i.e. storage.

ST already has this sort of stuff figured out.

> Also, NICs with hardware enhancements for protocol support,
> such as the Intel Gigabit Ethernet NIC have been very successful
> both in performance and market.  So, expect more there.

Intel provide no useful documentation so that is hard to evaluate.

> Finally, the RDMA option is targeted to allow use of standard
>  Internet protocols for SANs in place of  specialized protocols
> and networks such as  FibreChannel, especially for non-local 
> access.

So is ST and ST is proven technology. If you are going to break the protocol
to add hacks to it you might as well design the protocol properly based on the
past twenty years of learning where TCP is hard to get right. You need IP to
be compatible you don't need TCP.

If you want to solve the generic problem then you don't need RDMA anyway

Alan