|
SDN is a big deal, everyone thinks it's
going to make their lives easier. Eventually, but not without
pain. There is enough hype I'm delving heavily into this now.
Most of this concept of defined network is "contracts", this group of hosts to that group of hosts on x services. Hosts technically aren't in a broadcast domain anymore like they were, if you don't define the port-to-port flow as part of a contract policy, it doesn't happen. There in lies the problem - do you want to tell me every port connection every host needs explicitly for every application? Now you have to, as well as your bandwidth and latency requirements. Yes, really, and no "any" isn't good enough. As a network person, getting app people to define their networking requirements is about as impossible as pleasing them for the same reason. Do you know how to read the results of "netstat -anp"? There is lots of open or open-ish hardware out there now to do some very neat things right now. Openstack neutron to an opendaylight controller, running hardware switches in openflow mode (even cisco's can do openflow mode). There are a ton of open and closed source SDN controllers, Cisco, Arista, Juniper, really everyone has one, or has partnerships for one. VMware's NSX product is all about SDN flow controllers and distributed network features. As for hardware, commodity stuff is starting to be found cheap: http://www.ebay.com/itm/Accton-Edge-Core-Data-Center-Switch-AS5600-52X-48-port-10GBE-4x-40GBE-QSFP-/151754552764? Used, a thousand bucks, 48x 10gigabit ports, 4x 40gigabit ports, runs cumulus linux as the switch os. Anything that can do OpenNetworking (ON) support, which Dell, Quanta, and a number of other generic "white-box" providers do this now. Now almost every switch is based on broadcom chips, even cisco doesn't do much for custom silicon asics anymore, so there is a baseline compatibility with a lot of hardware now being all broadcom reference chipsets like that Accton through big names. Cisco and Arista's premier products are all broadcom based now, only difference is how they stitch the chips together internally, and the software. Facebook did this with the same broadcom chips themselves as with every random chinese company cloning the same approach from the bigwigs like that Accton and undercutting price, why not Facebook themselves? The final frontier for data center networking of who you buy will be more about the software api, and how expressive they can convey network hardware and transport concepts to non-networking developers, including global site concepts, vpn, load-balancing, ssl termination, dns, and extensive routing/switching. It doesn't remove complexity, it just makes it something else's problem, again. -mb On 07/25/2015 10:36 PM, Stephen Partington wrote:
|
--------------------------------------------------- PLUG-discuss mailing list - [email protected] To subscribe, unsubscribe, or to change your mail settings: http://lists.phxlinux.org/mailman/listinfo/plug-discuss
