IP networks have lately been under tremendous pressure to evolve. In Part 1, we reviewed some of the contributing factors that are making IP networks more challenging to manage including IPv4/IPv6, Gigabit Ethernet, WLAN, and the impact of new approaches to TCP.
With an eye to the future, in this column we'll continue the litany of influences including:
- Performance metrics for IP networks
- Edge controls
- NIC augmentation
- Security
- Machine-to-machine communications
Performance metrics
Driven hard by the needs of mission-critical VoIP, network and application managers have been working to define the relationship between IP network performance and the end-user experience. As described in a previous column (VoIP: The face of the new network police), VoIP has permanently raised the bar on network optimization -- simple, qualitative measures like MOS are now part of the everyday vocabulary.
However, it isn't stopping there. Both the ITU and the IETF are developing a series of standards that define IP performance and, as in the case of ITU Y.1541, specify performance requirements for categories of applications. These requirements provide the bases for network readiness assessment and for QoS mechanisms employed to support specific applications. Industry groups like the Application Index Alliance are promoting their own version of an application performance metric measuring user satisfaction. Each purports to define network performance in terms of end-user objectives and productivity.
Overall, this is will be a good thing. However, network engineers can expect to see a flurry of new "standards," measurement and monitoring features, and QoS mechanisms appearing in network hardware and software. More clearly defined by such metrics, IP networks are likely to achieve significant improvements in manageability and performance. And yet, engaging these emerging technologies will take some trial and error – and of course, there will some thrash associated with each new type of application as networks are forced to jump over evermore challenging optimization bars.
Edge controls
In the New Internet as anticipated by some groups (Fix before breaking), many of today's network woes will be resolved by the use of specialized network gear at the edges that provide critical middle-ware capabilities. The dominant core network vendors, Juniper and Cisco, have offered visions that claim to manage application performance from the IP network (dubbed InfraNet and AON respectively) with reference to future network infrastructures like Service-Oriented Architecture (SOA) and the Web-based Application Service Bus.
While the details are still sketchy, the intention seems clear: Aspects of application performance, security, federation, service negotiation, and the like, will be handled by these new network infrastructures. In the case of the traditional Layer 1-3 providers, this "pushes the network stack upwards" to include some functionalities that have typically resided in Layers 4-7 and been handled by the end-hosts. A typical benefit might see the enterprise customer of a network-dependent vendor (i.e. one whose product relies on the network) connect to some internal resource on the vendor's network via some automated federation mechanism implemented at the network edges.
Another obvious benefit is in end-to-end coherence of QoS mechanisms for specific applications (or per user). But as each of these "obvious benefits" come to live in the network layers, the complexity of the configuration increases. Mid-path network devices no longer simply store-and-forward packets – correct behavior will no longer be simply a matter of appropriate queue sizes and congestion mechanisms.
Mind you – firewalls already have given us an idea of what that kind of grief will be like. So maybe this will actually offer an improvement over the current level of complexity.
NIC augmentation
As network capacities move into the realm of 10 and 40 GigE, end-hosts will have increasing difficulty filling the pipe adequately. As described in Networking on the Edge (Networking on the edge), network interfaces are being augmented in various ways to deal with the growing disparity between CPU/bus and network capacities.
The typical means are TOEs and POEs (TCP and, more generally, Protocol-Offload-Engines) that reduce the work of the bus and CPU by pre-processing for Layers 3 and 4 on the interface itself. TOEs have also been used on 1 GigE networks but haven't caught on widely, primarily due to their cost. However, as the network capacities grow, the critical servers that need access to the full capacity will need some help -- and POEs seem like the likeliest candidates for high performance systems.
Other augmentations at the NIC level include multiple-port binding, aggregation/compression/acceleration, packet coalescence, jumbo frames (see Part 1), and other minor tricks that make a big difference in the right settings. Most often these augmentations are described by vendors as transparent; however, they regularly pose deployment challenges, in some cases resisting diagnosis when they are malfunctioning (transparency at its worst). Port binding for example has yet to fully mature as a technology and customers have reported difficulties. And jumbo frames are still waiting for a coherent solution to a mixed-MTU environment.
Security
The more you secure something, the harder it is to work on it. That's a truism that has its roots in all levels of technology. The front door key to our house or apartment has posed troubles for every one of us at one time or another. And so the simple act of sending, receiving, and routing packets had been complicated by the need to restrict access. Worse, even the simple best-effort feedback mechanisms of IP networks, ICMP, has regularly come under fire and been blocked or filtered on many networks.
Security fortunately, and unfortunately, comes first. So we are all too familiar with the need to support VPNs, encoding data payloads, securing files on network servers, and so on. There is a growing buzz surrounding phrases like "architecturally implemented security" that reflect InfraNet/AON influence. Again though, details are sketchy.
'Nough said. Moving on.
Machine-to-machine
Again, part of the service bus and SOA revolution (we're still waiting to see what survives the hype cycle), machine-to-machine communications will become a default part of everyone's networks. What isn't clear is how this kind of network traffic will scale over time. Some predict massive, even dominant, growth in machine-to-machine communications, particularly from various forms of telematics. They indicate that as all powered devices, from cars to toasters, are provided with wireless interfaces (such as Zigbee, 802.11a/b/g, Bluetooth, USB-on-UWB), networks will be flooded with data traffic. Some wireless networks today show on the order of 20-30% sensor traffic.
Complementary technologies such as RFID and GPS, as well as low-power sensors and sensor-network infrastructures, will provide all the necessary ingredients for the dramatic upsurge. Everyone and everything will be continuously linked and that sounds all very nice until you start go think about where all those packets are going.
And the impact on your networks? The usual -- more complexity required to manage yet-another-category of data traffic that is competing for bandwidth and impacting your IP performance. Only keeping track of the sources of the data will be like WLAN on steroids. OK, so that may be a matter of hyperbole – but it isn't clear how M2M is going to land or who it will affect.
Oh for those simple days of packets and routers, store and forward, Laurel and Hardy -- note that convergence wasn't even mentioned here. That was covered in another column (Making the triple play) but it shouldn't be overlooked.
Study hard. Be smart. Keep your head up and your eyes open. It's a jungle out there.
References:
ITU Y.1541
M2M Revolution
Application Index Alliance
10 GigE TOE
M2M article
Chief Scientist for Apparent Networks, Loki Jorgenson, PhD, has been active in computation, physics and mathematics, scientific visualization, and simulation for over 18 years. Trained in computational physics at Queen's and McGill universities, he has published in areas as diverse as philosophy, graphics, educational technologies, statistical mechanics, logic and number theory. Also, he acts as Adjunct Professor of Mathematics at Simon Fraser University where he co-founded the Center for Experimental and Constructive Mathematics (CECM). He has headed research in numerous academic projects from high-performance computing to digital publishing, working closely with private sector partners and government. At Apparent Networks Inc., Jorgenson leads network research in high performance, wireless, VoIP and other application performance, typically through practical collaboration with academic organizations and other thought leaders such as BCnet, Texas A&M, CANARIE, and Internet2. www.apparentnetworks.com
