I had a chance to listen into Paul Congdon's session at the #HPNetworkDay and captured some of the major enterprise IT trends he's thinking about that will influence networks over the next 3-5 years. Paul Congdon is a HP Fellow, who's a founding member of our CTO office, has been in networking for >25 years, and is leading figure in a number of networking standards activites.
Paul talked about 3 of the trends that are driving change in the networking space - XaaS, E2EV, and the future of convergence.
X as a Service - coordinated use of private/public/hybrid cloud, and driving a new service delivery architecture
A clear trend is the migration of applications into the cloud - whether it's a public, private or some hybrid combination, there's no doubt that cloud services will continue to grow in importance. For businesses, paying for applications by usage, running on flexible infrastructure; rather than having to buy the full infrastructure yourself allows the business to scale costs as the business changes. Perhaps most importantly, it allows IT to shift their focus from keeping the lights on to driving innovation for the business.
Many people have compared cloud to the mainframe model with the terminal as the browser, and the cloud providing a network of processors acting as the mainframe. This model doesn't take account of what you can do with the increasing level of intelligence at the edge of the network. Edge devices such as smart-phones, and laptops now have the processing capacity to do far more than the old terminal ever could - and this will allow us to improve service delivery by moving to distributed applications.
In order to understand what we mean - when you think about the combination of multiple public cloud providers and private clouds, we will each have to deal with multiple clouds for each service - which complicates the user experience. If you own a smart-phone - you can already see an example of this issue as you have to remember which application you used to capture a photograph, or make a note. The addition of the smart client at the delivery point allows the orchestration of those services and such that the user won't have to worry about where the applications are hosted.
If you take video for example, there's already a need for the network to understand what the user is doing - and what device they're using to optimize the stream. (Just take a look at the new vimeo universal player for an example of a smart application that was released today). This is about the service delivery agent recognizing what device you're on, and what bandwidth you have access to - then optimizing the service and pipe characteristics to meet those needs.
It can't however all be done from the device itself - because as the user changes devices - from say a Palm handheld back to his desktop PC - there is a need for the network to be able to coordinate the hand off of applications and streams between those devices.
Whilst this vision isn't here today, we already embed x86 processors into our edge switches (AllianceONE networking - services module) to create a virtualization environment - so that network centric applications can be moved to the edge of the network. You'll see HP talking more about the active edge of the network as we continue to define how applications, the cloud and the network interact.
End to End Virtualization (E2EV) - beyond just VMs, holistic virtual enablement
Virtualization in the data-center is clearly here to stay - with over 50% of workloads now running in a virtualized environment. The same drivers that exist for server virtualization (sharing resources across multiple applications), are just as relevant in the network within the data-center, and extend out into the rest of the enterprise.
In order to address the needs of cloud deployments, many vendors are creating virtual overlay networks through virtual switches to support multiple clients and networks running across a single virtualized infrastructure. Each of these vSwitches are now having to run as learning switches; and be managed as such by the network manager, creating additional complexity in the data-center. The use of general purpose processors for this purpose also steals processor cycles that could otherwise be doing more valuable work.
This has been necessary to this point, but it is incredibly inefficient. We need to flatten this layer, and move this back into the network, where we've got dedicated silicon that's designed for this purpose, whilst putting in the hooks that will allow the right level of interaction between the virtual machines, the vSwitch, and the network. You'll see more on this in the VEPA standards work that Paul is driving.
The key benefit of a new architecture to support end-to-end virtualization will be the ability to minimize the amount of change required to deploy new services, customers and virtual machines. The change should occur only at the edges and the core will need to support network isolation that goes well beyond the traditional 4K VLANs.
Convergence of Networking & I/O - truly converged unification of networking/storage/compute
We started out initially with racks of different boxes for networking, storage and servers, then moved to blade enclosures - initially with Ethernet and FC separately routed on the backplane. The subsequent generation of enclosures allowed us to converge this on to one network via Ethernet. This has reduced the cost of blades - but if you look at a blade - you'll still find HBAs and NICs.
In the future, we're working to drop even this part - and disaggregate the server; so that the server becomes just a bunch of CPUs, memory and I/O, all interconnected with a very high-speed, low-latency network. We're just about at the limit of what electronics can do in this space, but thankfully we've got HP Labs - who're making a huge investment in chip-level nano-photonics. Over time, this will allow us to use optics to interconnect chips directly.
This will gives us the ability to break apart the server - and allow us to 'render' a server for an application by grabbing a few CPUs, some memory, and I/O - and interconnect it with an optical network.
So rather than creating middleware that virtualizes the server to meet the needs of the application, we'll allow you to create a server to the exact specifications that need for your application. Paul calls this a 'truly converged infrastructure ' - a disaggregated server interconnected on demand with virtual I/O as needed, potentially saving huge amounts of power, and increasing performance at the same time.
To summarize, these trends are driving significant changes in the network. In the XaaS case, a new service delivery architecture involving an intelligent edge will appear. A new network architecture will evolve that efficiently provides network isolation beyond what we get from VLANs today to support end to end virtualization. Over time, the trend in convergence will lead us to a new system design that lets you dynamically configure hardware to render specific infrastructure for the application, not more inefficient middleware emulation software.
I think you'll agree that we've got some exciting developments cooking at HP - which will allow us to drive the networking and computing paradigm forward over the next few years.
We encourage you to share your comments on this post. Comments are moderated and will be reviewed and posted as promptly as possible during regular business hours.
To ensure your comment is published, please follow our community guidelines.
