Designing a Modern Fabric for Enterprise VDI with Plexxi
This post provides an outline as to why a Plexxi fabric is the best fabric to choose when deploying enterprise VDI. Before reading any further there are a few questions the reader should ask. Do I need better network? Am I ready for a modern network? If the answer was no to both questions, there is no need to read further. If the answer was yes and you would like to know what networking is like after legacy networking, then the following is for you.
The difference between a legacy network and a modern network is a modern network acts as a system that can be connected to flows and applications and thus be an enabler of business outcomes. When we think of legacy networks we think in terms of connectivity, but rarely do we think of networks as an enabler of business outcomes and that is because a legacy network was not designed to be a unified system – it was designed to be a distributed state machine. For decades this design served us well, but it also came with limitations. A legacy network is complex, it difficult to troubleshoot, it requires a staff of highly specialized people to support; it is static in terms of design and it is brittle which makes it expensive and inflexible.
|Legacy Networking – Technical Pillars||Modern Networking – Technical Pillars|
|Devices self-register, self-actualize||Devices operate as a system pool|
|Manage and configure devices||Manage and configure the system pool|
|Multi-tier design (ToR / Spine / Core)||Converged fabric / single tier design|
|Tune individual boxes and protocols||Controller tunes fabric to flows and apps|
|Distribute state / de-centralized scheme||Central logical control|
|Unknown fabric paths / rules / queues||Well known fabric paths / rules / queues|
|Inflexible bandwidth / static design||High utilization fabric / multi-path diversity|
|Variable time to failure handling||Fast failure handling / pre-computed|
|Wide latency variability||Known locality and deterministic latency|
Imagine we are a group of entrepreneurs pitching a business plan to a series of venture capitalists. It is doubtful we would get funded if we described our product set as being complex, closed and opaque that it required a team of people who were highly specialized, trained and certified by us to be successfully operated. That is exactly what we have been living with for close to thirty years for networking. It is almost as if we celebrate complexity. There have been efforts to overcome these obstacles and limitations, but these efforts have amounted to little more than incrementalism. A new protocol every few years for the same tried and true network design; a new operating system for network devices, but for use with the same network design using the same network protocols. A faster and bigger version of merchant silicon, but still the same network design running the same network protocols using the same tools for configuration and debugging. All of these efforts are little more than iterations on the past.
|Legacy Networking – Business Limitations||Modern Networking – Business Benefits|
|Focused on managing complexity||Focused on business outcomes|
|Domain of specialized skills and certifications||Domain of IT supporting the business|
|High Cost / Low Utility Return||Medium Cost / High Utility Return|
|Innovate around complexity||Innovate around the business needs|
|Network disconnected from apps and flows||Network connected to apps and flows|
When we talk to customers about networking and what life would be like after a Plexxi network, we make one simple promise: we promise to get networking out of your way. A Plexxi Fabric is designed to deliver the best possible experience for the users. In a Plexxi Fabric you tell the Controller what is important and the Controller tells you what the art of the possible is. That is the power of modern networking. The network engineer is not poking around a series of CLI windows trying to figure out what is going on in the network. The network engineer is not spending their time consulting third party apps trying to figure out how the network is configured. In the modern network, the network engineer designs the network as a system to support the needs of the business. At Plexxi, we are committed to moving the network out of the way of the business. If that seems like attractive proposition, then perhaps a Plexxi Fabric is for you.
Elements of a Plexxi Fabric
A Plexxi Fabric is composed of three elements: (i) switches, (ii) a Plexxi Controller and (iii) Plexxi Connect. Switches are typically off the shelf ODM designed white boxes. A Plexxi Controller is a fabric controller that runs on a VM. Plexxi Connect is an open source based event-based workflow automation tool. Together, these three elements form a powerful user-defined, open networking system. The keyword is system.
What Plexxi builds is a fabric as a service (FaaS) and this fabric is ideal for small-scale VDI deployments and excellent for large scale high-performance VDI environments. We take a common switch building block, in accompanying diagrams either a 3.2Tb/s Tomahawk or a 1.44 Tb/s Trident II white box switch and collapse spine layer into the switch. The switch becomes the access layer and the spine-switching layer. In modern networking, design practices use a high-performance controller architecture, engineers do not need to plan for a discreet switching layer often described as a spine layer or core layer.
We can build fabrics that are one dimension (1D) or two dimensions (2D), and these fabrics provide an immense amount of configurable paths (i.e. bandwidth) between end-points. The easiest way to think of a Plexxi Fabric design is to layout a grid (e.g. an aerial view of Manhattan) on a piece paper, roll it into a tube and connect the ends. This visualization tends to demystify the design concept. The design would also be completely useless if we ran spanning tree and other legacy fabric protocols, instead we use math.
Ethernet switching silicon is so powerful today and getting more powerful, that building discrete switching layers for fabric designs is really a hold over from era of networking design that is beginning to fade away. What is different about Plexxi is we can build this fabric and understand what is connected to the fabric (i.e. locality). By understanding where devices are located in fabric, and the multitude of paths between these end points, a Plexxi Controller can optimize the forwarding topologies in the fabric, a process we call fitting and enable the network to operate as a unified system controllable by the user.
Without spending a large portion of time on the technical details of a Plexxi Fabric, there is a default physical path construct and two controllers. One controller is located on the switch and is responsible for packet forwarding. The second controller is the logical fabric controller. It is within the logical fabric controller that the multi-commodity flow algorithms developed by Plexxi are deployed. In a Plexxi fabric the path of packets is never a mystery. The packet forwarding rules are known to the controller on the switch as well as the fabric controller. There is no mystery and network engineers do not need to have twenty CLI windows open trying to figure out forwarding paths or what spanning tree is doing. If this controller design architecture is foreign to the reader, a comparison illustration is provided that shows how Google architecture their fabric controller.
When the Plexxi Controller is connected to an application, the system can correlate between fabric end-points, topology paths and application workloads. This is the difference between a modern Plexxi network and a legacy network.
Building VDI Switching Fabrics Small and Large
In the following illustrations, the switch used is an ODM designed and manufactured ethernet switch. It is a 32×100 Tomahawk based white box switch or 1.44 Tb/s Trident II white box switch. You could take this switch and install any other switch OS of your choice and revert to a legacy network design. We tell prospects if you do not like the Plexxi network, feel free to choose any OS from any of the four or five network vendors offering a switch OS, as the hardware will never be wasted. To build a Plexxi network, we install the Plexxi OS using ONIE and then install the Controller on a VM.
The four switch design shown in the in the picture has fabric built into the switch. This is the single-tier, converged nature of the fabric wherein we turn traditional uplink ports into fabric ports and can dispense with the need for spine or core switch. A high-density Plexxi Fabric can be built on 10G or 25G or 100G paths that perform better without the need to design a spine layer. The bill of materials (BOM) for this design is four S3e switches and one Pod Switch Interconnect (PSI), which is a passive shuffle box. For such a small design, we only need a single LightRail dimension so the fabric is 2.4 Tb/s (600G x 4 S3e switches). The OSR of this design is 1.73:1 and it supports 10G, 25G, 40G, 50G and 100G clients.
5×10 Half Pod
Most of our customers who build fabrics that are 20 switches or less, typically build a single dimension fabric for cost reasons. For performance benefits, a second fabric dimension can be added at anytime, but for most designs the duality between surplus and scarcity becomes compelling as the fabric grows >20 switches. Greater than twenty switches is a good place for a multi-dimensional fabric. In this design we expanded the four-switch initial deployment to ten racks and added four additional rows of ten switches and a second LightRail dimension to the fabric. The actual fabric-wiring schema is not drawn accurately in this diagram. The X dimension is not a collection of five LightRail fabrics; it is really a single fabric dimension. The same is true for the Y dimension and we typically design the fabric with an offset between the X and Y fabric dimensions. Enough about wiring complex multi-dimensional fabrics. If you have a deep interest in how the fabric can be wired up and how the forwarding topologies can be manipulated, you can request a technical briefing.
The real beauty of the half pod design, it is the same design for a four rack, 160 server converged infrastructure (CI) design block. If you are enterprise IT person, the performance of the Plexxi Fabric can be had for as little as four racks or for twenty racks, or fifty racks, or one-hundred racks. With a Plexxi Fabric the IT cost/performance element becomes the converged infrastructure (CI) rack. With a legacy network design, the network was in the way of scaling out racks and rows of racks because it had to be engineered into the spine or into the core legacy network. With a Plexxi Fabric, the single-tier converged network design enables a true linear scale capability. This obviates the need for spine switches. Fabrics can now be built from a common 1U or 2U switch building block that brings benefits in terms of sparing, support, complexity and cost.
Legacy Network Design Evolution
A question we often hear is how to get started with modern networking? If the starting point is a multi-tier network design or classic fabric extender architecture how would one begin to upgrade the network? Do I have to upgrade all my server NICs and storage too? How disruptive will the upgrade be? Does it have to be a greenfield deployment for modern networking?
Let us start with the assumption that a customer has 16 deployed racks of compute/storage for VDI or just a generic network based on a popular legacy design approach using fabric extenders in a classic A/B network. There is no question that a new, clean set of sixteen racks would be fun for all to build and deploy, but sometimes that is not possible for cost and operational reasons. Is it possible to upgrade to a modern network, while keeping the legacy server and storage configurations in place? Is it possible to provide a path to the future of networking with 25G, 50G and 100G options when the time and budget are available? Those are some of the questions we often hear and addressed by the design outcomes illustrated in this example.
By replacing the traditional fabric extenders in each rack with a low-cost Trident II switches and placing Tomahawk based switches in the racks for storage and networking, a transformational modern network is deployed quickly. Instead of fabric extenders at the top of each rack, two 1.44 Tb/s switches are deployed, each of which has 240G of fabric uplink capacity, which provides each rack with 480G of uplink capacity into the fabric. This is a low cost, two-dimension (2D) fabric design. The compute fabric runs at 10G and the storage and networking fabric runs at 25G. The fabric optics are upgradable from 25G to 100G as the customer’s discretion.
By deploying a small number of Tomahawk switches to support storage and networking, the foundation is set to deploy new 25G, 50G and 100G converged infrastructure as needed. When the time is right, new compute and storage can be deployed by replacing the Trident based switches with Tomahawk based switches. In this design, the 25G LightRail would grow as new infrastructure is added and the 10G based LightRail would collapse, but the same design, cabling, software and operational procedures would be in place for 10G, 25G or 100G fabrics.
Benefits of Deploying VDI on a Plexxi Fabric
The organizational benefits of VDI are:
- Simplified desktop management
- Improved business scalability
- Centralized software control
- Increased security
- IT resource optimization
The downside to a VDI deployment is if it is not deployed on a network that can support the morning rush or compute intensive periods for high-end graphic users, the user VDI experience is less than satisfactory.
When you decide to deliver VDI on a Plexxi Fabric, you are providing the best possible network for your users. We have had customers realize reduced desktop load times by a factor of six (6), which are minutes to seconds. We no longer live in the era of 56k dialup modems; therefore we should not have enterprise class VDI that performs like the 1990s.
With a Plexxi Fabric the enterprise IT team delivers a unified desktop experience, with control, security and that is not encumbered by the silos of the past. IT organizations not only provide a powerful network for users, but they also lay the foundation to begin to use modern IT tools to deploy and manage IT services.