Jun 30, 2015
I recently attended the Open Networking Summit in Santa Clara. During a morning keynote session one of the speakers started to discuss his company’s SDN architecture, and made a comment on how that architecture is impacted by the fact that Moore’s Law is either dead, or soon will be dead. It was early in the day and I hadn’t had my second cup of coffee, but that comment still managed to get my attention. I started asking myself a bunch of questions, such as: Are we really seeing the end of Moore’s Law and if so, what does that mean?
According to Wikipedia, the law is named after Gordon E. Moore, co-founder of the Intel Corporation and Fairchild Semiconductor, whose 1965 paper described a doubling every year in the number of components per integrated circuit. In 1975, Moore revised the forecast doubling time to two years.
Many people think that Moore said that the performance/price ratio of semi-conductors would double every eighteen months. Actually it was David House, another Intel executive, who is credited with the statement that computer power would double every eighteen months.
Ok, I don’t believe every point made by every speaker at a conference. So why did this particular comment get my attention? The speaker was Amin Vahdat, who has a Ph.D. in computer science from Berkeley and who is a Fellow and Technical Lead for Networking at Google. According to his bio on Linkedin, “He has contributed to Google’s data center, wide area, edge/CDN, and cloud networking infrastructure, with a particular focus on driving vertical integration across large-scale compute, networking, and storage.” Two quick observations: Vahdat is a seriously bright guy and he works in the field of large-scale computing for Google, which is most likely the largest-scale computing done anywhere in the world. I don’t have the background to state a valid opinion as to whether or not Vahdat is correct about the death of Moore’s law, but given who he is and his position in the industry, you have to accept that he might be correct.
In a fascinating article published a few months ago, Gordon Moore said that “Moore’s Law” is not a law, but an observation and a projection. He also said that the current approach to making integrated circuitry, which is based on continually making things smaller and denser, is coming close to running into some fundamental limits, such as the speed of light. He added that there are other technologies that have been proposed to extend beyond what can currently be done with silicon, but he declined to speculate on how likely it was that they would be successful.
BTW, Moore and Vahdat are not the only people who think that Moore’s law has run its course. Just google a phrase like “the end of Moore’s law” and you will see lots of articles that support their view.
Economists have written extensively about their belief that Moore’s law has been a driving force of technological and social change, productivity, and economic growth throughout the late twentieth and early twenty-first centuries. If the economists are right, then the death of Moore’s law will have broad social and economic impact.
In addition to that broad social and economic impact, let’s look at the impact on computing budgets. Let’s assume the conventional view of Moore’s law that says that the performance/price ratio of computing doubles every eighteen months; now consider a hypothetical company that refreshes half of its servers every eighteen months. Because of Moore’s law, one option this company has is to spend the same amount of money on servers every eighteen months and have the processing power of those servers go up by a factor of three every three years. If Moore’s law is no longer valid, then this company’s three year budget for servers needs to triple if they want to continue to add the same amount of processing power.
Another area that will be impacted by the possible repeal of Moore’s law is the adoption of NFV. One of the key assumptions associated with implementing a growing range of functionality in software is that the general purpose computers on which this software runs will become increasingly powerful. If that doesn’t happen, then it will limit the range of functions that can successfully migrate from a hardware-centric implementation to a software-centric implementation.
Vahdat, Moore and others could be wrong and Moore’s law may have another ten or twenty year run. Alternatively, as Moore suggested, alternative techniques could provide the same kind of ongoing increase in the performance/price ratio of computing. However, there is a growing consensus that real laws, such as the speed of light, are causing Moore’s law to no longer be relevant and the possibility of viable alternative techniques being developed is just that – a possibility.
This would be a good time for IT organizations to spend some time thinking about how they would compensate for the repeal of Moore’s law. BTW, part of the Vahdat comment that got my attention was his statement that the Google architecture included increasing server performance by offloading functionality from the servers onto intelligent NICs.