Server Tech Predictions by Patrick Kerpan

I was recently asked by a trade magazine to prognosticate on the future of server technology from the CohesiveFT perspective. As the end of the year draws near, I thought I'd share the fruits of that labor with all of you. Enjoy! In addition to server market predictions, I also dabble in numerology and tarot cards in my spare time.

What will server tech look like in 10 years?

1. Server market size is going to grow dramatically faster than predicted.
   Over the last several years the data center server market grew about 6% per year in unit volumes. With the advent of virtualization and cloud computing some analysts are now predicting contraction in the server market; this as a result of people using virtualization for server consolidation projects. I don't agree. Maybe there will be a slight hiccup given the global economy - BUT - I actually think it is possible for the server market (volume sales) to more than double in the next decade.
   
   Here is why, enterprises are not limited in the number of problems they want to solve in order to manage and grow their businesses. They are not even necessarily limited by the money they are willing to spend to solve their problems. Enterprises are bound by their inability to manage the complexity of the IT infrastructure needed. Unchained from the complexities of IT management - enterprises are almost unbounded in their desire to bring IT resources to bear on their business challenges.
   
   As most of the significant challenges of data center management are offloaded to cloud and utility computing providers - businesses will be more than willing to buy massive amounts of computing, communication and storage resources to eliminate business growth obstacles.
   
   
2. "Grey market" / "Off brand" servers will present challenges to leading assembler/integrators.
   Cloud computing centers will be of such massive scale that the trend will be for them to move to cheap energy locations; which means away from traditional hosting centers - and potentially many of them in emerging economies, rather than in the US or Western Europe, where energy prices are high. As such - the suppliers to these computing centers may not be the traditional leading server vendors. As the millions of server units move "out of sight - out of mind" the faceplate and branding of the devices become less relevant; creating opportunity for emerging market hardware manufacturers to overcome the big brands. Fast, cheap, no frills, OK-energy efficiency will be the driving characteristics. Service contracts aren't the issue - plug them in and at some point, if it is dead, pull it out of the rack and throw away (like Google does).
   
   
3. Virtual Servers become the "Server" that matters - and there will be 250,000,000 new ones every year.
   As utility computing and cloud data centers become the dominant location for physical hardware servers, and as enterprises have a decreasing number of servers in their on-premise computing facilities, the Virtual Server (Virtual Machine Server) becomes the new focus of attention. It will be an enterprise's proprietary code and data processing embodied in these virtualized containers that is their source of business capability. Where those virtualized devices will run becomes a function of cost, reliability and latency; attributes of the cloud vendor, not attributes of the server hardware provider.
   
   As chips get even faster, more physical servers get sold, hypervisors get better, and enterprises become fully comfortable with virtualization it is easy to imagine 250,000,000 new virtual servers per year being assembled and deployed to physical servers. These become the new unit of enterprise productivity and the nexus of competition for customers by cloud computing centers and other value added service providers to this approach to computing.
   

Most promising brand new server technology?

1. Next wave server hardware refresh cycle makes virtualization the default deployment option; physical hardware deployments become the exception.
   With the next wave of hypervisors from Vmware, Citrix, Microsoft, Open Source Xen and KVM - combined with support for virtualized memory maps, virtualized IO channels (file and network) from Intel and AMD, the latencies introduced by virtualization will be under 1%. Given the advantages of virtualized infrastructure this new hardware will make virtualization the default. Today organizations use physical hardware by default, and virtulization is done after due consideration. With this next wave of hardware (subsequent generations to processors like the 16-core Intel® Xeon® Processor 7300 series processor and Intel Xeon Processor 5400 series), virtualization will be the default and physical hardware installs will only be done by extreme use-case exception.
   
   
2. Multi-core processors will grow in density (Intel/AMD) and in dedicated purpose cores (IBM/Toshiba Cell Processors)
   In the 5 year window more multi-core devices will become available. In many ways this will come as a form of "consumerization" where the unit volumes of home video game units will drive lower prices and more capabilities out of the multi-core market. These unit volume gains will allow for grey market server vendors to create cheap, powerful servers for utility computing.
   
   
3. Memristors will be part of controlling energy costs and capacity management in utility computing.
   The 7/8/08 edition of EETimes states, "In April, Hewlett Packard Laboratories researchers claimed to have 'discovered' memristors, which joined resistors, capacitors and inductors as the fourth passive circuit postulated by University of California at Berkeley professor Leon Chua in a 1971 paper." Commercialization of these devices will happen in this 10-year period as well leading to new forms of non-volatile random access memory (NVRAM) and potentially processor chip design as well. If only the NVRAM advances happen servers will be able to be for all intents "instant on and instant off" allowing fine grained power control across computing resource in utility computing centers.
   

What are the problems that need to be solved in server tech in the next 10 years?
Most of my thoughts on this is the mirror image of my comments above; many of the challenges are organizational and software-related including:

• Management of increased server units in the data center
• Reorganization of roles and processes to fully leverage the use of utility computing centers
• Migration of physical devices to cheap energy locations; installation of necessary bandwidth to cheap energy locations
• Need for instant-on, instant off devices
• Management of as many as 2.5 times one order of magnitude increase in the number of servers in the form of virtual servers
• Strategies for distributed a businesses computing infrastructure across multiple centers in order to avoid a catastrophic geographic failure
• New programming languages and frameworks may come into vogue to take advantage of mutli-core architectures; ERLANG (Ericsson's language for telephone switching devices is coming into fashion already today for these purposes)