competition. Although I could really see PCIe storage gaining traction
independently of EMC buying them.
talking about PCIe storage is a bit of a mistake IMO. simply because
it doesn't have any meaningful (predictive) definition. for instance,
one could imagine a PCIe device with incredibly unconstrained, near-raw
performance, with an onboard high-performance copy/management engine,
including many GB of decent-speed RAM to maintain read caches, small-write
coalescing and wear-levelling. perhaps that's what's actually being
talked about here. that approach is never going to be cheap, probably never
high-volume (commodity), or open-system-friendly.
but some people merely throw a few conventional SATA->flash channels onto
the pcie interface, sometimes layered with an off-the-shelf raid controller.
yuck.
I'm hoping for NVME, which is basically "more efficient AHCI for flash" -
it's still a block-level interface, but without all the historic constraints
of SATA (queue depth 32, less efficient register/command structure.)
my fear is that SATA is actually plenty fast for the mass market, so the
push to higher performance may never materialize. (I'd argue that's why
10G ethernet is happening so slowly.)
I confess I don't really understand how NVME is supposed to be packaged,
though - to be relevant to the mass market, it needs a standard pluggable
interface (like SATA)...
_______________________________________________
Beowulf mailing list, [email protected] sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
