Best talk on SDN I have watched Now i determined the subject of my thesis hope it was more than 30 mins .. even though this short period talk was very fruitful thank you sir :)
The key benefit is centralised AS network adminstration. This allows consistent network configuration (topology, authentication, provisioning) policies, consistent collection of network usage, a consistent analysis of network heath and more responsive ability to adapt to network challenges. Administering multiple dataplane resources requires a common view of dataplane match rule fields and actions. Challenges include performance bottlenecks and single point of failure.
"...we build a distributed system, and we are not good at that"@6:20- how's so? Last time I've checked, the Internet, which by the way is way more complicated than all these cool SDN powerpoint slides, works fine. So, from this example, we may conclude that the point quoted is false. Some other slides are also questionable. I liked the kinda (because it's still labbing) real examples though. Academically-intersting possibilities are huge for sure.
Asto508123 So? If you mean that It was not designed properly, this should make the job of "working" harder, not easier, shouldn't it? Yet it works. And there are perharps many thousands of man-hours of engineering and RnD put into it. So it was most definitely built. One could argue, perharps, that It "grew" to it's power due to the lack of centralized control. But ok, if the Internet is not a sufficient example, let's take a look at any big enterprize network. It was desighned and built, that we know for sure, by single and somewhat centralised entity. Somehow they manage to work, all these networks. They use BGP and OSPF/EIGRP, and maybe MPLS and other cool things. All pre-SDN. I'm not arguing here that we don't need SDN. The concept looks promising. But it's wrong to say that the networks we have today are broken: these networks work and fulfill customers/owners business needs at the moment.
Denis Borchev He never said that networks are broken. It's just not the case that there has been any proven to work slate clean method how to actually build a wide scalable network such as the Internet beforehand that was used. Designing and building something is different than let something grow wildly over the years. It also doesn't mean that it can't work, but it's probable that it doesn't work as efficient as it could; and that's exactly the case with big networks today. It's a horrible shore maintaining and scaling them and that's what SDN is addressing. You also mentioned BGP and other protocols and the keyword you used was "somehow". That's the problem. You don't want a "somehow". "Somehow" is the exact opposite of designing and building. I guess you "somehow" missed his point there.
Asto508123 "somehow" you are describing the problem and the point made better than he is. Thank you. Still, I can't see how SDN will make maintaining and scaling any less hard (well, in a network what was designed and properly built, it shouldn't be that hard, should it?). It looks more like moving workload around and abstracting (hiding) the work in the process. See, now we employ network engineers to set up the network, with SDN that would be programmers (with enough proficiency in networking) doing same job. Only the tools will change.
Denis Borchev A big advantage of SDN is the global view on the network. If you just need to define rules as those he mentioned in the presentation to set up your network, it is much easier, faster and, most important, more understandable what the network is actually doing for the administrator. Imagine you have hundreds of switches in your network and you want to enforce certain packets to take a specific path inside your network, all you needed to do in a SDN would be to write a rule to the maintaining Controller that is able to push your rule to every switch in your network that lies on the path. Let's say the length of this path is 20 hops, the classic way would be to configure each switch n on the path to forward the specific packets to switch n+1 single handedly. With SDN you just tell your Controller to do that work for you. It may be a pointless and simple example, but the real advantage shows when you want properties that need the network to dynamically adapt (like load balancing). If you are able to just define certain rules like "max 20% load for flow A on switch B, max 10% load for flow A on switch not B" that the Controller takes care of, then you would be totally lost if you wanted to configure that manually for hundreds of switches; and hope no external constraints are changing that would force you to change these rules again. Yes, it's "just" the tools that change, but it's such a big change that you can do the same (and much more) work in less time with much more flexibility and, again most important, actually understand the behavior of large networks without the necessity to dig deep into every shady corner of the network. The whole point of abstractions is to make complex things understandable that would otherwise be hard (or impossible) to grasp for human brains. SDN is just a logical step when faced with the inability to effectively manage and understand large networks. Just think big, really big, then you see why this approach is quite similar to why Software Engineering and OOP has emerged when Software grew bigger and traditional software development approaches became less and less efficient to tackle the problems that arose with developing complex software with a set of unsuitable tools.
There is a reason why he is a professor at Stanford, founded startups that have evolved in giant companies and is considered the worlds leading expert in Switch and router design and you are not.
total bullshit. i listened to the guy for 10mins, i still dont get a point what problems SDN trying to solve. he behaves more like industry evangelist than a professor. Talk is cheap, but his is expensive, VMW paid $1B for it.
Thanks for this talk, it was promoted on Coursera's SDN course.
I always like to hear talk from Nick, always informative!
Best talk on SDN I have watched
Now i determined the subject of my thesis
hope it was more than 30 mins .. even though this short period talk was very fruitful
thank you sir :)
Thanks for the MiniNet heads up in the video. I got the VM and run it in Vmware.
The key benefit is centralised AS network adminstration. This allows consistent
network configuration (topology, authentication, provisioning) policies, consistent
collection of network usage, a consistent analysis of network heath and more responsive ability to adapt to network challenges. Administering multiple dataplane
resources requires a common view of dataplane match rule fields and actions.
Challenges include performance bottlenecks and single point of failure.
bright future of networks from now on...
.
Nice tech talk ever. Thanks.
awesome!!!
It seems really great and so confusing at the same time.
Informative !!
Very Nice talk and informative
exciting stuff :)
Impressive!
Amm Good one
Now I understand what SDN is !
thank u
Good Talk!
Great :))
which giant evolved from his startupt? i know he is not one who founded Cisco。
Nicira
"...we build a distributed system, and we are not good at that"@6:20- how's so? Last time I've checked, the Internet, which by the way is way more complicated than all these cool SDN powerpoint slides, works fine. So, from this example, we may conclude that the point quoted is false. Some other slides are also questionable. I liked the kinda (because it's still labbing) real examples though. Academically-intersting possibilities are huge for sure.
The Internet wasn't built, it grew over the years.
Asto508123
So? If you mean that It was not designed properly, this should make the job of "working" harder, not easier, shouldn't it? Yet it works. And there are perharps many thousands of man-hours of engineering and RnD put into it. So it was most definitely built. One could argue, perharps, that It "grew" to it's power due to the lack of centralized control.
But ok, if the Internet is not a sufficient example, let's take a look at any big enterprize network. It was desighned and built, that we know for sure, by single and somewhat centralised entity. Somehow they manage to work, all these networks. They use BGP and OSPF/EIGRP, and maybe MPLS and other cool things. All pre-SDN.
I'm not arguing here that we don't need SDN. The concept looks promising. But it's wrong to say that the networks we have today are broken: these networks work and fulfill customers/owners business needs at the moment.
Denis Borchev
He never said that networks are broken.
It's just not the case that there has been any proven to work slate clean method how to actually build a wide scalable network such as the Internet beforehand that was used. Designing and building something is different than let something grow wildly over the years. It also doesn't mean that it can't work, but it's probable that it doesn't work as efficient as it could; and that's exactly the case with big networks today. It's a horrible shore maintaining and scaling them and that's what SDN is addressing.
You also mentioned BGP and other protocols and the keyword you used was "somehow". That's the problem. You don't want a "somehow". "Somehow" is the exact opposite of designing and building.
I guess you "somehow" missed his point there.
Asto508123 "somehow" you are describing the problem and the point made better than he is. Thank you.
Still, I can't see how SDN will make maintaining and scaling any less hard (well, in a network what was designed and properly built, it shouldn't be that hard, should it?). It looks more like moving workload around and abstracting (hiding) the work in the process.
See, now we employ network engineers to set up the network, with SDN that would be programmers (with enough proficiency in networking) doing same job. Only the tools will change.
Denis Borchev
A big advantage of SDN is the global view on the network.
If you just need to define rules as those he mentioned in the presentation to set up your network, it is much easier, faster and, most important, more understandable what the network is actually doing for the administrator.
Imagine you have hundreds of switches in your network and you want to enforce certain packets to take a specific path inside your network, all you needed to do in a SDN would be to write a rule to the maintaining Controller that is able to push your rule to every switch in your network that lies on the path.
Let's say the length of this path is 20 hops, the classic way would be to configure each switch n on the path to forward the specific packets to switch n+1 single handedly.
With SDN you just tell your Controller to do that work for you.
It may be a pointless and simple example, but the real advantage shows when you want properties that need the network to dynamically adapt (like load balancing). If you are able to just define certain rules like "max 20% load for flow A on switch B, max 10% load for flow A on switch not B" that the Controller takes care of, then you would be totally lost if you wanted to configure that manually for hundreds of switches; and hope no external constraints are changing that would force you to change these rules again.
Yes, it's "just" the tools that change, but it's such a big change that you can do the same (and much more) work in less time with much more flexibility and, again most important, actually understand the behavior of large networks without the necessity to dig deep into every shady corner of the network. The whole point of abstractions is to make complex things understandable that would otherwise be hard (or impossible) to grasp for human brains. SDN is just a logical step when faced with the inability to effectively manage and understand large networks.
Just think big, really big, then you see why this approach is quite similar to why Software Engineering and OOP has emerged when Software grew bigger and traditional software development approaches became less and less efficient to tackle the problems that arose with developing complex software with a set of unsuitable tools.
There is a reason why he is a professor at Stanford, founded startups that have evolved in giant companies and is considered the worlds leading expert in Switch and router design and you are not.
total bullshit. i listened to the guy for 10mins, i still dont get a point what problems SDN trying to solve. he behaves more like industry evangelist than a professor. Talk is cheap, but his is expensive, VMW paid $1B for it.