Traffic Flows Through the ACI Fabric

hi my name is Jo on a sec I'm a technical marketing engineer with AI NS bu and I'm here today to discuss the ACI forwarding operations with you give you a little bit of an idea of how we forward traffic within the ACI fabric so to kick off what I've drawn here on the board is a spine leaf architecture known as a class architecture this is the base architecture of how we design and deliver a CI fabrics there's a lot of advantages to this architecture one it promotes the promotes the traffic east to west which is the way datacenter traffic patterns are moving today rather than north-south traffic patterns we bazan designing in the past it also offers us very linear scale with this type of fabric the class architecture spine leaf design when you need more server or device connectivity ports you simply add a leaf and you can add leafs up to the scale of the amount of spines that you have in the fabric when you need more redundancy or more paths for bandwidth within the fabric you simply add a spine switch we typically connect every leaf to every spine and the spines will only connect the leaves everything else in your network architecture is going to connect up here at the leaf layer within the ACI model we look at this with two different spaces of where we where we're looking at traffic we have the infrastructure space which exists here where we're forwarding traffic out to what we call the user space and the user space can consist of a single organization or if you were talking about scaling out to provide infrastructure-as-a-service services we could scale out to 64 thousand tenants sitting within this user space so anything external to the fabric would be what we'd consider that user space connected up at our leaves you'll see that we have several things that would make sense at your access layer of your data center in a three-tier design like our server ports and our server hosts for virtual machines we also are going to connect up our service devices those could be virtual or physical service devices shown here as load balancers and firewalls in a physical format or virtual format down here and we also connect up our external networks now when we connect up these external networks we don't look at a CI in isolation of what you already have in your data center you've already got an investment in data center switching infrastructure and we connect and integrate cohesively with that infrastructure and that's shown here whatever networking gear you choose to have in that space will exist in cohabitate with a CI from that user space we also have our connectivity to our outbound networks so this could be intranet shown here in green this could also be internets shown here in blue so anything external so basically you would take the router that's serving that traffic you would connect it up at the leaf layer of the ACI fabric now when we talk about fabric forwarding inside the infrastructure space we utilize VX LAN or virtual extension extensible land this is becoming the industry to facto standard for a layer 3 overlay topology this is prevalent in all modern ship sets including merchants silicon platforms and many of the products that Cisco offers what a layer 3 topology offers us is the ability to abandon spanning tree in the constraints of the spanning tree protocol and have a fully routed robust multimeric that doesn't have to rely on blocking lengths to prevent loops so we do this using layer 3 routing and then we can encapsulate that traffic from that user space into the X land and then use that VX land overlay to be able to provide layer 2 adjacency when we need to so we can emulate that layer 2 adjacency you would get out of a VLAN while giving you the extensibility of the X land the scalability of the excellent and then several other advantages that we offer utilizing when traffic comes in to the infrastructure from the user space that traffic can be untagged frames it can be 802 dot1q VLAN tag frames it can be VX LAN it can also be env GRE so these are two overlay technologies that are prevalent in the data center what we want to do is take any of this traffic and normalize it within the fabric so when it's received from this virtual machine at my leaf shown here what I'm going to do is translate that into VX land within our infrastructure space and then transport it to where the destination is going to be I can transport that let's say for instance these are Microsoft hyper-v servers in using env GRE I can take that env GRE encapsulate into VX land transfer or dit to the leaf it needs to exit on and maybe these would be VMware servers using the excellent I can rien capsulate it into a VX land frame basically drawing that connection between two different frame formats across the fabric we can do this between any hypervisors workloads or any physical device whether that be a physical bare metal server or a physical service device for layer four through seven devices so this gives us the ability to completely normalize the traffic coming in rather than double encapsulate it here within the infrastructure space I'm going to re encapsulate it a line rate and I'm going to translate that back out into the frame format that's required at the exit or egress port we're going to do the same thing providing routes between devices within this VX LAN fabric as well as providing their external routes to either that intranet that internet or out to the existing data center infrastructure when we look at connecting that data center infrastructure into a CI what we do is allow in here either typical subnets on any given V RF or a VLAN from any given device externally we then translate that into the fabric as external entities or external groups that can become parts of the application centered infrastructure that we use building out our logical model when we translate fabric in the when we translate traffic in the fabric the excellent provides us quite a few capabilities as the first one I mentioned was is it allows us to alleviate spanning tree protocol but it also allows us to get location independence within the fabric an IP address itself is intended to identify that device for forwarding purposes but it's become very tied to the physical locality of that device and the way in which we use it so within our fabric what we do is take where a device exists or what a device's addresses its IP and we map that to a VX land virtual network ID or vina in which is going to help us identify where that is at a given time so what this means is if I'm talking about this virtual machine as VM 1 this virtual machine is identified by its IP address within the server and the V need or leaf it's sitting on right now if this virtual machine were to migrate over here and this virtual machine was now vm1 what happens is I translate that vide in into the new location so that now I know this IP is that same device existing at a different location this allows me to provide very efficient forwarding to a device while still allowing the flexibility provided by workload mobility so this gives us a very robust fabric extremely scalable and allows us to handle normalization of forwarding from our user space across the infrastructure space to any given end point when we look at this the scalability of the fabric is based on both the way in which we designed it and this spine leaf design that we built out so we get linear scale from both a performance and a cost perspective from a very small scale 100 or so ports at a very cost-effective price to up to a hundred thousand 10 gig ports and a million endpoints so this is the architecture and the user space forwarding for how a CI operates we have several other videos to talk about the way in which our logical model operates and we'll have videos coming to talk about how the fabric itself handles forwarding internally so please stay tuned and go take a look at our other videos thank you very much

Channel: Cisco

Views: 27,865

Keywords: data center, virtualization Fabric mode, ACI Fabric, Insieme Fabric, Cisco Application Policy Infrastructure controller, APIC, Nexus 9000, Nexus 9500, Nexus 9300, Leaf Node, Spine-Leaf architecture, spine node, promise theory, cisco

Id: Z9eZYM7J33s

Length: 8min 13sec (493 seconds)

Published: Tue Apr 22 2014