SDNRG Agenda IETF 90 Thursday, July 24, 2014 0900-1130 EDT (Canadian) ================================================ CHAIR(s): David Meyer Nick Feamster AGENDA o Administriva 5 minutes - Mailing list: https://www.irtf.org/mailman/listinfo/sdn - Web: http://trac.tools.ietf.org/group/irtf/trac/wiki/sdnrg - Scribes? the first is about terminology, nordmark suggested this o Agenda Bashing 5 minutes David Meyer skiped o SDN Layers and Architecture Terminology 30 minutes draft-haleplidis-sdnrg-layer-terminology-05.txt Evangelos Haleplidis changes since london added more peer-reviewed literature + correlation with RINA architecture one question on mailing list was why do you make the distinction control vs management? Help to make a better design choice, and the management plane is a part of SDN that has been ignored a bit by the community => they created a subsection in the draft give now the reason why (timescale, persistence, locality)k [Yakov]: you summed up very well assuming no partitioning. You prove that people call SDN a control-plane they don’t understand. SDN is [Kostas]: before management presented as control plane [Diego] so big discussions (almost religious) on previous IETF about control and management. You found a very elegant way to tackle this [Dave] point out “no-size-fits-all” we try to hit the optimal surface, if you disagree we hope it will be minor [Dirk] nice paper “CAP for network” from berkley and KTH. [author] we have referenced it [Yakov] not to the mic, can’t ear the operational plane is not really a plane, there is no communication, but we decided to keep it as a plane to be considered equal to the other so to be able to group things. So see “plane” it as a collection of “features" ask for comment [yakov] do we really need another plane? Services providers have their business services, below the operational … the management plane can be devided in many parts if you really want to have all, it will big a huge doc. So what do you want [kostas] we don’t try to capture what is now but we also place ourselves in the frame of NFV, but what we could see later [yakov] I didn’t know I was going that way. if you go to NFV, you need to use orchestration then. The orchestration is put on top of everything [kostas] you are talking about another draft that we should then write together [yakov] we could say “orchestration is behind the scope of this document” NFV discussion [yakov] provlem I have with that. I am ok to have a management plane about NFV but what if you want to extend to virtual functions and application functions? Then some of these function will help making the networking. If you open up this, it is very far away from the SDN, it is more about computing. I believe we should not go there. [author] it is to point out the kind of interface to use NFV, the NFV will use this interface [yakov] yeah but other things may use this! as long as it is short it is ok but keep it small [author] it is a very small paragraph [diego] it is reality but it is not a plane, this is stretching a bit too much the boundaries. For software functionality you need another layer. Points not addressed decided not to include more SDO, that should be documented on the wirki but not more we decided to not discuss research issues, use the wiki for that we were asked to create a matrix of SDN project but as it changes all the time, not necessary [yakov] do we really need this wiki? There is already many forums about SDN [kostas] we do not propose to make a wiki, we just say that this should not be in a draft that will anyway be frozen one way or the other. A wiki is appropriate in a sense that it is a moving site We believe it is mature enough to move to publication [dave] Lars can you give us advises [lars] unlike at the IETF where we need consensus, IRTF does not need as long as there is no conflict. [dave] I want to publish it so we don’t circle around and around and around [yakov] we don’t have contentious left [yakov] except terminology we are kind of agreeing [ed] need to add a statement in the document that orchestration is not [lars] it is clear that SDN and NFV are not underplayed, there is overlapping but they are different. IRTF is anyway more liberal than IETF [dave] we will make the changes suggested here and try to submit for publication o Design and Implementation of an OpenFlow Hardware Abstraction Layer 20 minutes Kostas Pentikousis presentation alien project. Small project on the continuation of ophelia that can be seen as similar to gini how good would it be if tomorrow ISP would have only OpenFlow in their network How would it look like? A big mess with many technologies => need to abstract the hardware (the technology)! The idea is to be able to deal with large array of devices. interesting point is to look at software reusability Architecture is very close to what has been presented i previous draft, but not exactly the same, we want to go beyond OpenFlow The big blue background is about the hardware abstraction. Two layers: the cross hardware and the hardware layers In our technical reports it is shown how it is implemented As ophelia is virtualised we needed to support virtualisation There is a discovery part, when a device comes up, it must “register” to the cross hardware layer Presents the implementation, plenty of different hardware, they all use the same key code base some advertisement on where it has been demonstrated and where it will be presented [yakov] you call this an open flow HAL. How bound to OpenFlow is it? How much would change? OF is fundamentally not modular. How much of this is a concept of how to do [kostas] the software use does not come from the code but the velocity of features that we want to be able to account easily for. Remember that it is a research project. The code reuse is that you have the cross-hardware. It is re-use of “drivers” somehow. Not application [kostas] why open flow and not something else? OF [ed] you seem to relegate hw specific function to netconf [kostas] netconf is for the cross-hw layer [ed] how much do you think you would be able to move to new versions [kostas] moving out of 1.0 is hard for many reasons, so with this it should help [ed] migration between management and control [kostas] out of the scope of the project it is easy to install :) o OF-DPA (OpenFlow Data Plane Abstraction) 20 minutes Joseph Tardo non research project: it is about implementations free to use and download reminders on SDN principles plenty of choices to implement hardware switches [yakov] it is not futuristic, it is already available [author] who? [yakov] let’s talk about that offline [uri] can you define slow vs fast? [author] pps (CPU slow, hard fast) thousands in difference [uri] we can have discussion about those numbers implementation compatible 1.3.1 running on broadcom asics [yakov] do you do group table in hardware [author] yes we do [yakov] talks about a problem (I don’t understand) [dave] how many are aware about this problem? (which one I dont know…) [adrew ? alcatel] ( I did not understood what he said) open hardware and software implementation links for github [yakov] each flow table has a single field in the example, is that a limitation [author] of course we have action list and action set [jamar] would have been more useful if not bounded to OF [jamar] do you have performance figures? [author] I don’t know [jamar] what perf you have for table updates [author] I don’t know [jamar] why not in the linux kernel, why the SDK? [author] to communicate with the chips [jamar] would work with a driver [author] yes [ the alcatel guy] (could not follow this have to check offline) o NEMO Language (NEtwork MOdeling Language) 15 minutes draft-xia-sdnrg-nemo-language-00 draft-xia-sdnrg-service-description-language-00 Susan Hares what is missing between app and network wen we want to do networking device interface is not network interface! [???] can’t hear [author] [dave] does NCL has a formal description [author] yes but the guys are still checking to be sure formal and code are the same [author] we will submit it to the mailing list later o Alpha fair traffic engineering using SDN 15 minutes Peter Ashwood-Smith [dave] why does that need to be convexe? [author] I can’t cover that here, trust me they are but I don’t have time to explain conclusion : can be computed very fast on NetFPGA to globally optimise the network [yakov] in many cases you cannot preempt flows [author] I am aware of that problem [luyan] why does has to be combined with segment routing? [author] does not need, but I am biased against source routing :) [luyan] how many labels can you handles between millions nodes? [author] that implies a large amount of state [author] that does not work with lose routing, I must be able to bind to a precise link [luyan] then you will have a lot of segments and so many labels on the stack of packets [xiaoqing] how do you rate allocation accross the flows? [author] it is not every single possible path between route and destination but the good ones. How do you place the flows on good paths [xiaoqing] if you consider it as a SDN? why in hard? [author] you need a mechanism to deal with very high performance vector if you have a general purpose harware able to do vectorial stuff it can do the job, but I don’t know suck hardware. We looked at GPU but moved to netfpga. We tried multi-processors but memory exchanges between cores were a problem [xiaoqing] how to enforce all the flow rates to all applications? [author] the output of the decision is input of traffic policers of course you have some mismatch between the optimal and the real but if you scale up the number of flows it is closer and closer to the optimal o Cooperating Layered Architecture for SDN (CLAS) 15 minutes draft-contreras-sdnrg-layered-sdn-00 Luis Miguel Contreras Murillo [?] is this work from telefonica A+D [author] yes it is o Autonomics and SDN - Complementary Concepts 10 minutes Michael Behringer [yakov] I like the first few slides (to be completed) [author] we have a running implementation. We just need link local (we use IPv6). Yes it is IP, but it is done automatically [yakov] with sensors they are not SDN but they have to bring back the data over multiple hops and that connects them to the SDN world o Application-based Network Operations (ABNO) framework 15 minutes http://5g-ppp.eu/wp-content/uploads/2014/04/TOUCAN.pdf Daniel King [uma] do you have any prototype? [author] wait a couple of slides [yakov] (a joke that I did not understood)