Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on August 27, 2014 3:38 pm wrote:
> You know Intel used to make the claim that the ARM ISA is not suitable to browse the internet.
> That sounds ridiculous now given there are far more ARM devices than x86 browsing the internet...
> But here you effectively say "ARM is not suitable for IO, you need x86 for that"...
>
Its not like it isn't common knowledge that Intel has the best SATA parts on the market or that the majority of I/O hardware is designed for Intel. Or that the majority of device driver development is generally done for x86 and the x86 drivers are more robust than for other hardware.


> The fact is none of this is in any way ISA related. The Marvell cores used in the cheapo NAS units are quite
> low-end compared to modern SoCs - some of them are still XScale derivatives. If someone wants to make a mid-end
> ARM-based NAS with a modern ARM core or wider/faster SATA then that is certainly possible. The software stack
> can be recompiled for ARM due to being open source, so that's not an issue either. I'm sure you agree Cavium
> knows a little about NAS servers, and I bet Thunder X will end up in their high-end versions.
>

Cavium overall knows very little about NAS servers. They have at best a very very small market in NAS of any type. The low end, mid end, and high end of the NAS/SAN markets are all completely dominated by x86. The low end in particular used to be dominated by MIPS/ARM then ARM alone and is now pretty much entirely controlled by x86.

And no, it isn't ISA related except in so far as the ecosystems related to ISA and the companies supporting various ISAs. The vast majority of the ARM ecosystem doesn't do I/O and has minimal to no device support. Now that may change, but it will take quite a while.