[Info-vax] VMS QuickSpecs

[Stephen Hoffman]

On 2015-08-17 12:15:17 +0000, johnson.eric at gmail.com said:

> On Friday, August 14, 2015 at 8:18:43 AM UTC-4, Dirk Munk wrote:
> 
>> The ethernet protocol for FCoE is far less robust as FC, FCoE didn't 
>> make it, iSCSI adds the overhead and latency of the IP stack, in most  
>> situations we don't need it.
>> 
>> FC still is technically superior to anything ethernet can offer at the moment.
> 
> Rather than argue in a traditional fashion, I'm curious to see which of 
> the following statements you'd agree with.
> 
> a) There are some storage problems that only FC is equipped to deal with
> 
> b) Ethernet based solutions can be an appropriate solution for smaller domains
> 
> c) In general, the ethernet solutions would be called "good enough"
> 
> d) The ethernet solutions will have a lower upfront cost than their FC 
> counterparts
> 
> e) Providers of ethernet based solutions will grow at a rate faster than FC
> 
> f) In five years time, FC will be even more of a niche product, and 
> ethernet based solutions will be the dominate commodity of choice for 
> everyone.
> 
> g) In five years time, the number of problems that is true for (a) will 
> have shrunk


Here's some data... For pricing. Intel dual-port 10 GbE PCIe NIC is 
~$220.  Intel single-port 40 GbE NIC is ~$480.  (Those are capable of 
various protocol offloads, as well.)   For comparison, HP-branded dual 
8GbFC FC HBA is ~$456.  ATTO 16GbFC is on sale for ~$1625.   Those are 
street price, quantity one, right now, via Amazon.

Next-iteration predictions... Formal 32 GbFC prediction was ~2016.  The 
prediction for the 400GbE standard (IEEE P802.3bs 400GbE) is ~2017, per 
a working group timeline.

Here's some opinion...   Outboard FC is increasingly squeezed between 
inboard non-volatile storage, outboard Ethernet where the low-end gets 
up to 40GbE, and outboard InfiniBand.   That's before any discussions 
here around trying to add support to OpenVMS FC for what Ethernet can 
already do — IP — and before getting FC HBAs and FC cabling replicated 
everywhere, and before discussing the savings from having one set of 
wiring and devices, and not two parallel sets.

<http://www.intel.com/content/dam/www/public/us/en/documents/product-briefs/ethernet-xl710-brief.pdf> 

<https://en.wikipedia.org/wiki/3D_XPoint>

As for what outboard storage will have to contend with, 
block-addressable flash is available now, and Intel is claiming 2016 
for volume 3D Xpoint production.  For outboard storage, start stacking 
up the 16 GbFC HBAs or the 40GbE NICs, or 100 GbE NICs when and where 
you can get those to meet latency and bandwidth — and where you have 
the switch ports for it.  All of the major switch vendors have had 100 
GbE ports available for a while, too.  This to be able to keep up with 
faster storage, and to remain semi-competitive with inboard 
non-volatile storage and with designs based on redundant arrays of 
servers — in OpenVMS terms, think host-based volume shadowing (HBVS, 
RAID-1, mirroring) over three-node clusters interconnected via 40GbE or 
100GbE NICs 3D Xpoint, with no outboard storage required.

Then briefly ponder what byte-addressable non-volatile storage (akin to 
3D Xpoint) means for the minimally-available 48-bit physical address 
space on most x86-64 boxes, and to the baroque block-addressable 
outboard storage designs being held over from the rotating-rust 
block-addressable era.  Also ponder what byte-addressability means for 
the present HBVS "disk" abstraction, too.  Maybe a switch to RDMA via 
Ethernet or InfiniBand.   Then what's involved to more than a few 
applications, if you no longer have to use $qio or $io_perform to 
access non-volatile storage.  Or for that matter, potentially 
addressing and executing code directly out of non-volatile memory, with 
RAM caching.

Can't really see the point of implementing IP over FC with the way 
things certainly seem headed.  If that's not already obvious.



-- 
Pure Personal Opinion | HoffmanLabs LLC