What do Virtualization and Cloud executives think about 2011?  Find out in this VMblog.com series exclusive.

The "science" of technology prognostication is actually quite predictable...there are a set of well established and proven projections for base technologies. One can reasonably assert that these projections are, in fact, both inevitable and inexorable. Examples include the increasing ubiquity of 10 Gbps Ethernet as a mere precursor to 40/100 Gbps Ethernet. Semiconductor fabrication technology is rapidly moving from 65nm, essentially skipping 45nm, onward to 28nm and beyond. Sometimes, however, linear extrapolation doesn't always work. A good example of this would be the reluctant realization several years ago that simply increasing the frequency of single-core processors was merely a recipe for exponentially generating heat. The result is that multicore processors and virtual machines are now the architecture de rigueur.

Ultimately, the Achilles heel in any multiprocessor architecture is the approach adopted to address the well-chronicled memory latency problem. While the vendors offering general-purpose cores often cite the merits of the specific implementation of their multilevel cache architecture, they demur from talking about the (very) real-life need to statefully process millions of flows at 10 Gbps (and soon 100 Gbps). Stateful processing such large numbers of flows at rates of 10 Gbps and above can only be attained by using a highly-threaded architecture! No matter how sophisticated the cache, its utilization will have tumbled to a hit-rate of less than 20 percent under such conditions.

We now live in the Wikileaks era, which is spreading advanced cognitive dissonance among fragile-egoʼed politicians worldwide, live an avian flu pandemic. Many commentators suggest that there are no more secrets. Unless one believes that Utopia is just around the corner, it might be prudent to get some data encryption.

Thus, my predictions for 2011 are:

1. Intelligent multi-threaded pre-processors will become an integral part of the data center architecture and will be used for a variety of I/O functions including maintenance- of-state for a large number (millions) of flows, intelligently directing traffic to the appropriate general-purpose core or virtual machine, and performing traditional networking functions such as protocol header parsing, classification and traffic management. Besides the obvious solution to the memory latency problem, small, intelligent multi-threaded pre-processors which are specifically optimized are also much more power efficient than general-purpose cores for the pre-processing function.

2. The top-of-rack model will shift from an appliance form factor to a much greater level of integration, achievable through the same multi-threaded pre-processors equipped with a large number of I/O ports.

3. While everyone is familiar with Metcalfe's Law, it might be wise to also heed Metcalfeʼs Axiom, which is "never bet against Ethernet". Thus, the historical debate about converged networks and the nuanced positioning of alternative architectures such as Infiniband and Fibre Channel may yet continue for some time. The fact is that Ethernet will inevitably win. 2011 will see the emergence of a new class of clustering technology such as the Torus clustering alternative to Fat Tree networks. The initial implementations will use 10 Gbps Ethernet, but it is only a matter of time before 40/100 Gbps implementations will emerge.

4. Security will become mandatory and thus ubiquitous in all of the manifestations defined under the umbrella term "security" (i.e. encryption in-flight, at-rest, authentication, intrusion prevention (IPS), extrusion detection, DDOS prevention, legal intercept, forensics, etc.). This presents great growth opportunities for some of the leading companies such as Sourcefire and McAfee.

About the Author

Jim Finnegan has over 20 years of senior management experience in the networks and communications business. His career has included extended periods based in Ireland, the UK and Santa Clara, CA, where he now resides.

Jim has had a career-long commitment to the development of an engineering culture that expects first-pass success using disciplined software and silicon development methodologies. This focus has most recently been demonstrated in Intel Corporation where he oversaw the development of Intel's entire portfolio of network processors, several of which went into production in first-pass silicon. Jim began his career at Miles-33 in the UK and subsequently worked in senior engineering and management positions with Digital Equipment Corporation, Tellabs and Racal Data Group. Following Racal, he built the engineering organization at Basis Communications and was a key member of the executive team that negotiated the acquisition of Basis by Intel Corporation in 2000. At Intel, Jim became General Manager of both the Network Processor Division and the Communication Infrastructure Group's Technology Office.

Jim has Bachelors and Masters Degrees in Electronic Engineering from The Queens University, Belfast.