There's a lot of hype here. Most of what they're claiming to be their product is just the direction of the datacenter industry. The idea of routing L2 over L3 (VXLAN, OTV, NVGRE) and new hypervisor switches isn't earth shattering, simply a new set of ideas that are being created in many places at the same time.
This specific product really only works in environments that are using linux based hypervisors (Xen/KVM/Openstack) in a datacenter environment. It doesn't address the typical campus network, or the more common ESX/ESXi/HyperV virtualization environments. The major block for most vswitches is VMware's lack of flexibility for other hypervisor network switches. The only option is the Nexus 1000V, and it's comparable to the dVs. The problem with both is that it requires Enterprise plus licensing which is roughly twice the cost of Standard.
In short, the difference between Nicira and the 1000V using VXLAN is not large. They both have hypervisors independent of switching fabric that allow large scale L2 flexibility. The only difference is hypervisor availability. It may be a while until we see VMware loosen up on their hypervisor, and equally time until things like the 1000V make it to Xen/Openstack.
Those links are very useful. However, I note that at each refuted point, the author is constantly being surprised by an actor creating a solution that does indeed do something new. This includes interoperability with VMWare hypervisors and userland networking performance with virtual networks.
Given the above and that most networking hardware is still configured by hardware engineers, this sector does seem ripe for disruption.
I'm a networking guy, and wouldn't consider myself a hardware person by any means. Sure, I'm not a coder, but I don't think it's accurate to call it hardware. Very little network engineering is hardware specific. Hardware goes bad, and there are cabling issues, but those problems will exist as long as networks have wires.
The challenging portion of configuring networks is keeping track of all the stacks and protocols. MAC, IP, virtual L2 features, virtual L3 features, virtual L2 with L3 features, routing virtualization, L2 loop detection/prevention, L3 loop detection/prevention, maximizing bandwidth without adding complexity or cost, multicast delivery strategy, broadcast/multicast control, best practices, and integration the application/TCP layer as well. I would say nearly all of those problems exist with or without something like OpenFlow, and it's already software. I love this part about networking, as every layer in the stack is modular and can be virtualized, so you're continually keeping track of protocols on top of protocols, and it all evolves so quickly. To say network engineers are no longer necessary after you virtualize networking is analogous to trying to cut 90% of server support staff once you're 90% virtualized. The added complexity may just compensate for the lack of hardware.
The other option is once the virtualization density is so high that virtual routers become feasible, and virtualized switches have more value, that you start to see 'hardware' go away. There are some very large cloud providers that are at this scale and can benefit, but the everyday enterprise isn't quite there yet. That's why these technologies are still a few years from mainstream, but they will definitely catch on.
Outside the datacenter, and even for the data center core, I think we'll continue to see customized silicon as opposed to commodity computing processors pushing the network traffic. Even the 'software' Google switches are on something like the Broadcom chip set, specific to network forwarding. Cisco's one of the few companies that develop their own chip set, which is one of the points of contention in the article, but they continue to find reasons to justify the cost.
Ugh. I hate this kind of article. It hypes a new technology without bothering to explain how it works or even what it does. It might as well be a tweet:
Martin Casado is hellasmart. Nicira is the VMware of networking.
Check out Open vSwitch http://openvswitch.org/
There's a startup to be done that will make IP itself meaningless; what we need is a generalized, bidirectional overlay datagram service (think: Akamai, but for packets, not files --- or, Akamai crossed with Skype) with abstracted addressing. We tried to build this in '99 but failed for want of a killer app for the service (also, we failed because our VCs installed management that felt our best bet was to take our series A and try to compete directly with Akamai by building colos).
This is, I think, the long term answer to both network neutrality and the IPv4 address crunch; IP could be relegated to a similar long-term role as Ethernet.
Sure; "Content Centric Networking" or "Content Based Addressing" is, I think, a TIBCO term, which I think stems from Cheriton's TRIAD project at Stanford.
> “The truth is, in 10 years, you’re not going to have highly skilled, highly paid people working with networking hardware.”
I'm skeptical.
PC/Server virtualization didn't make IT go away, instead it made the job more complex. Now you need someone to run the VMware cluster and understand the nuances of virtualization. You still have a bunch of servers to patch and backup. And you still need the help desk to fix the same problems we've always had (printer doesn't work, my internet is down).
However, you gain a lot of flexibility and agility (which is the point of the article).
I've just not seen new technology reduce _technology_ jobs. In fact its usually the opposite. We need to coin some kind of catchy "law meme" like Greenspun's tenth rule, Conway's, Sturgeon's etc.
You're right. Typically, greater investments will always follow improvements in efficiency, due to the improved ROI. If it suddenly becomes easy to deploy geographically dispersed, multi-tenant networks that are easier to manage the result will be more multi-tenant, geographically dispersed networks.
Highly skilled people may not be working with networking hardware per say, but they will certainly be doing networking.
Hardware is the important word there. It's just commoditization in action. Like if a hard drive fails in your RAID array you just swap it out, it doesn't mean there isn't a well-paid sysadmin operating the LVM that sits on top of it. Or a node in your compute farm.
I really wonder how the plan to virtualize all the network equipment.
With VMWare (and others) you had a fixed set of OS'es that mostly assumed a homogenous x86 platform to run on.
But this seems to be the opposite. You have a (very) heterogeneous physical layer and with a homogeneous software layer.
Anyone have any details on this?
I work at a telco with very heterogeneous network, we have to run a lot of separate and expensive (b/c of oracle/solaris licensing) management software to make this all work. So this is obviously interesting to me.
"Nicira, the most intriguing startup in Silicon Valley". Really? It's not clear why a virtual switching company is the most intriguing but I guess if Wired says so...
The title may be a little hyperbolic, but if you read the article, the implications of what Nicira is working towards are indeed quite impressive -- and not just because "Wired says so".
Software networking is clearly the future. It is probably long term, but complex hardware switches which do not use commodity processors and commodity software languages will definitely fall by the wayside, even if they are as big and bad as Cisco. Vyatta was a first attempt at a software router, but this is really the next step at software switches.
Hmm some times custom hardware is needed for switching just of the top of my head you not going to impliment CAM on a bog standard hardware are you.
I could imagine some custom FPGA based modules that you could reprogram on the fly when you needed to regonfigure your networking hardware I am not sure what that buys you networking quipment doesnt have the scale that say a standard pc does apart from maybe acess level switches.
At this moment this seems to be more of a point and drool interface for those that cant hack IOS or pass the CCIE exam.
There's some interesting (one might say verging on dishonest) terminology here. An incredibly complex and specialized Broadcom switch chip is defined as "commodity" while a very similar Cisco switch chip is defined as "proprietary". What is being commented on is really business models (horizontal vs. vertical integration), not technical aspects.
This sounds like bad news for freedom of the internet. The easier it is to code up filtering, blocking, detection, etc, the harder it is to have a free internet.
Open Networking Summit (http://opennetsummit.org/) is a three day conference just for SDN/OpenFlow. If you're interested in seeing what companies are doing with OpenFlow, you should check it out.
if the one thing this article does is take you to http://www.openflow.org/ and get you to read some things, I think it's served enough of a purpose. openflow is really neat and will let you do things with your network that you probably think should already be possible.
This specific product really only works in environments that are using linux based hypervisors (Xen/KVM/Openstack) in a datacenter environment. It doesn't address the typical campus network, or the more common ESX/ESXi/HyperV virtualization environments. The major block for most vswitches is VMware's lack of flexibility for other hypervisor network switches. The only option is the Nexus 1000V, and it's comparable to the dVs. The problem with both is that it requires Enterprise plus licensing which is roughly twice the cost of Standard.
In short, the difference between Nicira and the 1000V using VXLAN is not large. They both have hypervisors independent of switching fabric that allow large scale L2 flexibility. The only difference is hypervisor availability. It may be a while until we see VMware loosen up on their hypervisor, and equally time until things like the 1000V make it to Xen/Openstack.
Here's a better breakdown of what's actually happening for the networking/virtualization curious: http://blog.ioshints.info/2012/02/nicira-open-vswitch-inside... http://blog.ioshints.info/2012/02/nicira-uncloaked.html