Container networking is on the agenda, which should be really interesting, given how Cisco is seemingly so tightly welded to its hardware when containers are almost completely abstracted from the hardware. Networking containers is going to become a big deal, because of two major trends (that I just wrote about for an upcoming issue of CRN Australia incidentally): the decoupling of networking hardware and software, and the rise of automation and orchestration.
Cisco is already moving in the software-defined direction with UCS and ACI, but it’s still centered in hardware. There are loads of startups that are working on the “switches are just servers with lots of Broadcom Trident II chips” approach backed by the Open Compute Project and ONIE. Just like we’ve seen in server land, the purchase of hardware is separate from the software selection. Linux and Windows will both run on a variety of x86 based hardware from a variety of vendors, and that fact has spawned a huge number of startups doing software and HCI type things, not least of which is VMware.
And as we saw with the decoupling of operating system from hardware that virtualisation brought us, we’re starting to see virtual networking operating systems pop up. I expect to see container-based version of the idea as well. Imagine if a firewall config change was a rebuild/recompile and deploy the way Docker applications are done today? What about a BGP route-reflector?
We’re also going to see a sprawl in container-like entities an order of magnitude worse than what we have now with virtual machines. They’re small and designed to be deployed en-masse. Of course we’re going to see loads of them sitting out there doing whatever it is they do, and they’ll all need to be networked somehow. The only way to cope with the sheer volume will be through automation, because humans just can’t handle that sort of scope cost-effectively, and we’re already seeing IT staff-to-device ratios come way down.
This is as it should be, because manually updating ACLs or routing tables entries is boring and humans are bad at it. I’m still somewhat agog at how long it took the networking world to ditch telnet for ssh, and even then the CLI continues to rule supreme when it’s a tedious and error-prone way to configure hundreds of devices. I recall using Tcl/Tk script to automate MPLS VPN rollouts back 15 years ago (also the Java-based hell of Cisco’s VPNSC product, but let’s not go there) so why oh why isn’t everything REST API based already?
Happily Cisco have some modern tales to tell here, as they acquired cloud management software startup CliQr not long ago. That’s also on the agenda for a chat, and I really want to dig into this more, since I’d just heard of CliQr before the acquisition.
There’s also Metapod, Cisco’s converged insfrastructure/cloud-in-a-box version of OpenStack, so that could well be interesting, not least to get a handle on how people are deploying OpenStack in the enterprise, and how it links into existing systems.
How Cisco is going to pack all this into their session will be a challenge, but I look forward to it.