[inline] Sent from my iPhone
On Jun 29, 2017, at 21:06, Kenneth Seefried via vcf-midatlantic <vcf-midatlantic@lists.vintagecomputerfederation.org> wrote:
From: Dave McGuire <mcguire@neurotica.com>
I didn't say it was impossible, I said it was pretty tough to do.
And I said it wasn't. It wasn't in the old days if you knew what you were doing and it definitely isn't now in the 95% of everything is Ethernet/IP world.
We've all seen those interfaces, and those of us (myself included) who have used them have run up against the same old problem of driver availability, OS version dependencies, etc etc.
Some of us were and are capable of dealing with those issue and routinely do.
In fact the first widely-deployed router that was capable of routing IP on a T3 at wire speed while handling BGP was an IBM RS/6000 7012-320 with a synchronous serial card in it, bundled by IBM as the 6611 Network Processor.
I'm quite familiar with the 6611. Awful things to build large networks with (I did) unless you really, really needed SNA/APPN weirdness that it took Cisco a while to figure out (I, regrettably, did). But, yes, commodity hardware. Lot's of IBM network network gear is "IBM [RT,PC,PS/2,Power] + ARTIC card" under the cover.
That said, I believe a properly configured Wellfleet BLN could handle T3 routing + BGP before the 6611 (in 1991?). The Wellfleet BCN certainly could, but I think production BCNs were delayed till after the 6611 shipped. I was building out a worldwide Wellfleet-based network at the time (1100 routers?), but some of the details have faded. Wellfleet had the worst user interface ever conceived (SiteManager aka SiteMangler), but we deployed tons, including some interesting FDDI->FDDI routing scenarios (so the bandwidth was there). Let's not stroll into the minefield that was Wellfleets OSPF implementation....
Those were heady days.
But this is not the norm by any stretch.
Perhaps not the norm by Cisco sales numbers, but that's kinda like saying *BSD doesn't exist because Microsoft.
- IBM sold a lot of 6611s. For a long time. - Internet IMPs were Honeywell DDP-516s - we've been doing routing on commodity hardware for a long time. - Nokia sold tens of millions of bog-standard rackmount ATX PCs (later CPCI) running FreeBSD (relabled IPSO) as firewall/routers with optional V.35/X.21, HSSI, ATM & FDDI interfaces. I still use a Nokia/Sagnoma sync serial card from an IP440 in my lab. - Most of the earlier Juniper line are repackaged PCs running FreeBSD (nee JunOS). - SnapRoute is a Linux-based routing platform that AT&T has vetted for 5G network routing on commodity hardware (http://about.att.com/story/white_box_collaboration.html). - etc., etc., etc.
I've used snaproute - it's a lot of Go-language based bits atop OpenNetworkLinux. I'm hoping they certified a more tested and debugged/in-house version compared to what I ran a few months ago. There was this nice inability to ... create trunk switchports. Attempting to would completely destroy the config and I couldn't figure out which subproject to file a bug against.
And that ignores the whole Software Defined Network (SDN) trend which is predominately OSS+White Box Hardware. Little guys like Google, Amazon & Facebook are betting on that.
There's still a bunch of ASIC handling offload of forwarding so I'm still confused as to what precisely is software defined in SDN ;) Aside from, say, BigSwitch which does at least have a neat centralised control/management plane.
The fact that it can be done, and was very occasionally done, isn't really the point.
Sure, when 'the point' becomes cherry-picking whatever allows you think declare yourself 'right'. Par for the course, as they say.