Q&A from our NFV Webcast with Mirantis and Metaswitch

We recently hosted Mirantis Director of Product Management for NFV Sutapa Bansal and Metaswitch VP of Product Marketing Micaela Guihat for a webinar on deploying NFV with OpenStack.  We had way more questions than we had time to answer, so in case we didn’t get to yours, here are all of the questions and answers for that day (including those we answered on the call.)

(For more information, you can check out the webinar for yourself.)

Q. What are typical use cases for NFV?

Sutapa Bansal: There are around 400 VNFs that can be deployed on top of an NFV platform, but the most typical use cases are around evolved packet core, IMN, that is IP multimedia subsystem, and virtual CPE and SBC. There was this notion that either in an enterprise or in a residential use case you would get a particular appliance and that is you’ll be limited with the life cycle of that appliance. But, now with virtual CPE as well as many elements, such as firewall, load balancer, DPI and many others, you can actually upgrade each of these services independently, and there’s far more flexibility to deploy this in a more agile manner. So, these are typical use cases that we see.

Micaela Guihat:  I think what we are seeing in the industry is that the Session Border Controllers (SBCs) have become really the first application for virtualization for service providers and I think there are real good reasons for that. One is the fact that it’s very easily bounded as a network element, so it allows a very low risk testing environment, but also it can prove quite beneficial as a new business model enabler.

So, we are seeing quite a few new business models being put in place — things like specific virtual SBCs to customers, being part of the virtual CPE that Sutapa just mentioned now, and also very clear business improvements from an economic perspective in terms of CAPEX savings as well as OPEX savings in the long run. So, very much an easy thing to try as a first network virtualization function.

Q. What is the adoption rate for NFV architecture among telcos today?

MG: We are seeing actually quite a bit of live deployment as I said, so I would say that the rate of adoption is the highest that I’ve ever seen from a new architecture perspective, even if you look at technologies that you probably have forgotten existed, such as intelligent networks or even IMS. That took a very, very long time, but if you look at the virtualization environment, we started talking about this a couple of years ago, and we are already seeing live deployments in the network.

From our perspective, we have over 20 deployments live or being commissioned right now, mostly in tier ones, but also tier threes for instance, or tier twos. We have Telecom Alaska for one of the examples and in that virtualization environment, it’s extremely beneficial for them in that particular geography. So, bottom line is the carriers are moving very fast and really demanding knowledge from their partner vendors.

Q. What is the difference between NFV and VNF?

SB:  NFV is network function virtualization. It is the end-to-end solution which encompasses NFV infrastructure layers. It encompasses VNFs, and then it also encompasses the management and orchestration layers. So, it is the end-to-end stack that we showed that consisted of the management and orchestration (MANO) layer, consisting of VNFs and the underlying infrastructure.

Now, VNF is a part of this solution, and VNF stands for the virtual network functions. It is the individual network functions. You classify them into two sets of network functions.

One is the telco VNFs, the telco virtual network function such as IMS, EPC and likes of those. The second is the more traditional network functions, such as firewalls. VNF is the actual network function that is the functionality that is being used.

From the NFV service, for example, a virtual CPE service may actually lead to the plumbing of multiple VNFs, such as the firewall, router, DPI to deliver that packet.

Nick Chase:  So, basically, NFV is the composition of multiple VNFs into a single system?

MG: Yes.

Q. What makes SDN different from NFV functionality?

SB: SDN is software defined networking, so if we look at the overall NFV stack, it has the infrastructure layer as one of the components and SDN is one of the ways to implement networking as part of the infrastructure there.

There are many solutions, like open source solutions such as OpenDaylight, and many vendor solutions such as from Contrail (or its open source cousin, OpenContrail) or ACI or Big Switch, which provide advanced networking capabilities on top of your regular infrastructure. So, for example, if you look at OpenStack. it has the Open Virtualization Format (OVF) as the default auto switching layer, but then instead of OVF you can use, for example, OpenContrail to provide additional service chaining capability when you want to pull all these VNFs together.

Q. Is this something we can test on a couple of servers and switches like we deploy Neutron SDN? Hope the webinar will cover hardware and software requirements. Thanks.

SB: It’s part of the initial Mirantis and Metaswitch testing, so essentially you would deploy OpenStack on a standard networking and standard set of servers, and on top of that you deploy the SBC as a VNF based on the Metaswitch deployment guide, which is available on their website, so it’s a very standardized offering which can be run on a standard server.

Having said that, if you want to benefit from the additional NFV features such as SIROV, then to use those features you may need SIROV capable NICs such as Intel 82599 or CORPA or other NICs.

MG: Right. But, I want to just add that the virtual deployments are really now cookie cutter with SIROV or a native virtual switch, and yes, you have to know what you can use and what you need to use, so for us it’s absolutely all available as a cookie cutter, and we have the deployment guides.

Q. Do we have some numbers for how many sessions, external networks an vSBC can handle?

MG:  That’s a very good question, and I will be more than happy to answer specific capacity questions if you drop me an e-mail, but, overall it depends on what type of deployment and how many virtual CPUs you want to use. So, really scalability is not necessarily the issue. It’s the computing power as well as how you want it deployed.

Again, there are certain – I wouldn’t say limitations, but there are certain architectures in which you may want to use a hybrid environment for a certain transcoding capability. In some cases, software transcoding using virtualization is perfectly fine. So, it’s a hard question to answer without a specific example of requirements on CPU as well as requirements on the architecture.

Q. How do you see OPNFV and the OpenStack telco group working together?

SB: That’s a very interesting question. We can see there are some vendors who are active in both the groups, but then there’s another set of vendors who are more active in the telco working group, and yet another set of people who are working in OPNFV, so the idea with both these groups is to actually ensure that the telco and NFV requirements are truly represented in the OpenStack community, and I think that is the chance we are taking in moving ahead.

MG: I completely agree. Absolutely we need to make sure the carrier’s interests are being represented and I would encourage everybody to participate as much as possible in both.

Q. Do you plan to support service chaining?

SB: Service chaining is essentially the idea of ensuring that if you deploy multiple instances of these VNFs, the packets will flow in a particular order. So, for example, in an SGI LAN environment where you have multiple instances of these components including ADC and SBC, but you want, let’s say, your traffic to flow through the firewall and then to your ADC.

So, service chaining allows you to do that to ensure the typical order for packets. What makes it interesting is that it should be done on a dynamic basis based on the actual packets, and the information inside that packet header, and not just something which is done in a static manner.

The SDN partners that we have, they do support service chaining as well as their SDN offering and there are also open source implementations.

Q. Is SRIOV implemented, or is it a roadmap item?

SB: It’s shipping, and it’s a capability that you can use today, but we’re working on making it easier to use.

Q. Can NFV insfrastructure co-exist with conventional IaaS infrastructure?

MG: Absolutely. From an SBC perspective we are seeing quite a bit of movement towards initially a hybrid type of deployment, because you will do pockets of virtualization, and then depending on the degree of urgency, I guess, for each carrier the transition will take shorter or longer term.

One of the things that I wanted to mention that Sutapa also mentioned earlier is the fact that when you’re looking at deployment, these type of virtual functions, they all need to be agnostic of the orchestrator that you’re going to use, so in general they need to be agnostic in making sure that they all blend together well in the environment that you have chosen as a carrier.

Q. Does NFV dictate the SDN is used? Or can traditional networking be used?

SB: NFV doesn’t dictate that you have to use SDN if you can’t – it is also based on the use case. So, for example, if you’re just deploying a virtual firewall or virtual ADC on top of an OpenStack deployment, there is no real need for you to have advanced SDN mechanics.

But, let’s say you’re deploying a virtual CPE and you’re deploying from many different vendors and not as an integrated solution, then you need a mechanism where you have the ability to do service chaining, and you have the option to use SDN as a way to use service training, or you need to use other mechanism to provide that functionality. So, it is a choice. It is not something that is mandatory.

Q. What is DPDK, and is it being shipped or is it just experimental right now?

SB:   Well, essentially it is a set of libraries known as Data Plane Development Kit. So, it is a set of libraries which ensures that you have higher performance, and it works with OVS, so with OVS 2.4 it became available on 21st August, and essentially what that means is that Intel did a fork of OVS some time back, and now there is OVS mainstream which is supported with DPDK. Once you enable DPDK, you can use multiple cores to deliver functionality and you have higher performance.

And these are all just of part of the NFV features, SRIOV, DPDK and the other one I think we didn’t get to is Hugepages as well. So, these are a set of capabilities which enable you to get higher performance for the VNFs.

Q. In my opinion, using SRIOV capabilities has some limitations for cloud capabilities. Since you seem to support the use of SRIOV could you shed some light on pros and cons?

MG:  Yes, there are some developing needs for media, and in particular the denial of service scalability that needs to be still worked on SRIOV as well as the accelerated vSwitches. On the other hand, we are doing this today, and for the deployments that we are seeing, it’s quite manageable and it’s doing its job.d

Q. Which OpenStack edition do you support?

SB: We are shipping Mirantis OpenStack 7.0, aligned with the Kilo release, and we are aligned with the community release.

Q. Do we really need an actual router to communicate between control plane and virtual services?

MG: In order to communicate between a cloud service and an external device you normally use an actual router. However, it is also possible to put cloud VMs onto the same layer 2 network as bare metal devices, in which case you may be able to communicate with them without a router.

Q. Do we have a demo?

MG: We’re working on one – watch this space!

Q. Can you provide a list of customer (names)?

NC: Many of our deals are covered by NDAs, but Michaela and Sutapa both mentioned some within the webinar.

Q. Can you please briefly talk about Metaswitch’s “Project Calico”?

NC:  Check out the OpenStack:Now podcast with Metaswitch’s Christopher Liljenstolpe.

Q. Is hot-plugging supported with sriov networks?

MG: Not yet. See https://blueprints.launchpad.net/nova/+spec/sriov-interface-attach-detach.

Q. Are Cisco and VMware SDN partners?

SB: Yes.

Q. How does the vSBC integrate with Cisco UCM?

MG: Please contact Metaswitch directly for more information here.

Q. Is the Intel 82599 the only NIC vendor you have qualified/support for SR-IOV/DPDK?

SB: Besides Intel, we also support Mellanox. We will be adding support for many more.

Q. Does Mirantis support the whole solution, including the OpenStack implementation?

SB: Yes, contact Mirantis directly for more detailed information here.

Q. Are you looking at VAS services such as SMSC Voicemail systems etc. also to be VNFs?

SB: Yes, contact Mirantis directly for more detailed information here.

Q. Can you please describe more details about scale out and scale up in the slides?

MG: Please contact Metaswitch directly for more detailed information here.

Q. What is the hypervisor of choice in a Mirantis/Metaswitch deployment?


Q. What about NFV mano? How do you address its features (which products)?

MG: The vSBC has an Orchestration API for use by a VNF Manager. This API supports creation and initial configuration of instances, monitoring, scaling and operational automation.

Q. What is the role of OpenStack in an IoT or wearable environment? Can software on wearables be virtualized?

MG: OpenStack clouds can host services used by wearable apps. Wearables are not an obvious place to do virtualization, but maybe it will happen one day.

Q. How do I set integration with sigtran or ss7?

MG: By peering the vSBC with a PSTN gateway.

Q. Are you planning to develop SDN controllers as well? and probably integrate them with an NFV solution?

SB: We are looking into the best integrated NFV and SDN solution that works with choice of solutions.

Q. Are you providing cloud platform only for NFV of you have some reusable frame work as a “partner ecosystem”?

SB: Yes we have an OpenStack NFV ecosystem partner ecosystem.

Q. Do you offer an HA & GR solution?

SB: Yes we support HA, and GR is in discussion. Contact Mirantis directly for more detailed information here.

Q. Is Live migration working on a VM with sRIOV or DPDK?

MG: Not yet. See https://blueprints.launchpad.net/nova/+spec/sriov-live-migration.

Q. Also, how about service orchestration in a multi-cloud environment, including public and private clouds?

MG: This is possible using the Mirantis / Metaswitch solution as a component. We are not providing the overall service orchestrator as part of our solution.

Of course there was much more to the webinar than we covered just in the Q&A.  Please join us for the webinar replay.

Subscribe to Our Newsletter

Latest Tweets

Suggested Content

Mirantis Cloud Platform
Machine Learning in the Datacenter