Pages: pp. 6-7
In "The Real Cost of a CPU Hour" (Apr. 2009, pp. 35-41), Edward Walker proposes a "modeling tool that can quantitatively compare the cost of leasing CPU time from online services (cloud) to that of purchasing and using a server cluster of equivalent capability."
This article cites the $0.10 per CPU pricing from Amazon and later makes a comparison with the costs of purchasing hardware. However, from my view as someone working in a company that offers outsourcing services, I presume that the $0.10 per CPU pricing from Amazon includes other important costs like the staff who provide the support, the monitoring infrastructure, and all the other resources needed to provide an acceptable service level. For that reason, I think the comparison is unfair since the data for the "purchase" option does not include these costs, and the "lease" always includes them. I understand that these costs also exist for an organization that buys the hardware, but they could be significantly different depending on the organization's efficiency. In my experience, these differences could easily be on the order of 10 to 15 percent.
The article describes good tools for calculating and comparing the costs related to hardware, but an organization must consider the "total cost of ownership" to be able to compare with the cloud or any outsourced offering. The article provides good input for estimating this TCO, but the organization must look at its overall operating costs, which are significant.
Martial Van Neste
The author responds:
It is absolutely true that we must consider the "total cost of ownership" to fairly compare with the cloud or any outsourced offering. In fact, the model I propose allows you to incorporate the suggested support, infrastructure, and other costs associated with providing an acceptable level of service in the "purchase" case.
My model assumes an annual amortized cash flow (profit-cost) CT for each year, T, a purchased cluster is in operation. An IT organization then simply needs to estimate this cost in deriving a comparative analysis.
To see how this is done, the first example in the article (NSF Track 2 HPC cluster) uses $7 million in the annualized-operations cost (including infrastructure, cooling, and support personnel) for the purchase case in determining the true cost of a CPU hour. For the second example (a compute blade rack cluster), we only assume the electric utility cost in the annualized-operations cost in the purchase case. We do not explicitly incorporate any support personnel cost because we assume an IT organization will be able to leverage existing IT administrators to support the small cluster. For larger clusters, we presume an organization will need to estimate its support costs, as we do in the first example, in its comparative analysis.
Thank you for making a great observation, as well as presenting an opportunity for a good teaching moment.
Thanks for publishing the thought-provoking "The Robot Scientist Adam" (Ross D. King, et al., Aug. 2009, pp. 46-54).
It's one thing to deduce a logical truth from a set of facts but quite another to come up with a new idea. The ability to generate a new idea is unique to man. According to Aristotle/Aquinas, the ability to produce a new idea is the quality that distinguishes man from other animals. And now we have a machine that can perform that very function.
In his Phenomenology of Science, Thomas Kuhn offered an in-depth discussion about how humans come up with new ideas. Kuhn discussed "normal science," that is, the formulation of hypotheses, then the gathering of data, then statistical tests that may or may not prove of benefit. Normal science iterates around the experimental space, looking for clues, and eventually exhausting all possibilities. The experimenter digests the results of the iterations. He is essentially out of ideas. At this point—the point of imminent discovery—he moves in a different direction in terms of his perspective on the problem. He has a new path to investigate. And this eventually leads to the new paradigm, which can displace existing knowledge on the subject.
Being an avid reader of Sherlock Holmes as a youngster, I am glad to see that his unique ability to formulate new hypotheses has been captured in computer software. Holmes always got his man.
I am hoping that someone can come up with a way to step back from a problem, gain a new perspective, and then iteratively investigate as Kuhn described. Perhaps I'm quibbling over words here, but I suspect that Kuhn pointed to something more in his discussion—the ability to think creatively as humans do.
The authors respond:
We thank Dick Brodine for his positive comments regarding our article. However, we feel compelled to reply to his repeating the view of the philosopher Thomas Kuhn that there are two types of science: normal and revolutionary. This idea may be helpful in explaining certain historical events, but we feel that it does not accurately reflect the actual scientific process, nor is it helpful in designing robot scientists. We prefer the analogy with chess, where there is a continuum in player skill from novice to grand master, and where computers slowly improved with advances in computer hardware and software until they now play at the world championship level. There is a similar continuum in the ability to do science, from what robot scientists can do today, through what most human scientists can achieve, up to the level of a Darwin or Newton. We predict that the abilities of robot scientists will improve with advances in computer science, and there will be no discontinuity in the type of problems they can deal with.
The first sentence of the IEEE's "About Us" statement says, "A non-profit organization, the IEEE is the world's leading professional association for the advancement of technology."
International conferences are a vital tool for experts, visionaries, professors, and students to get together and share their ideas and dreams, helping to advance technology and share knowledge. However, I note that recently the expenses for attending an IEEE conference have increased to the degree that I cannot afford to do so.
Take as an example a 2009 IEEE conference for which the early-bird registration fee is $822 for an IEEE member and $1,050 for a nonmember. The hotel cost is $189 per night, and the conference lasts for five days. Thus, the total cost just for the registration and the lodging would be $1,767. Adding the airfare and incidental expenses, the total cost for me to attend this conference would be about a month's salary. In some countries, this cost could be the equivalent of two or three months' salary. As a result, it is nearly impossible to attend without financial support.
Things were not always so expensive. Ten years ago, a typical registration fee was about $400. Our salaries and the consumer price index did not double in the past 10 years. What is the reason for increasing the registration fee year after year? Soon, conferences will become a rich man's club. Only people who live in countries with deep pockets will be able to afford to attend. The internationality and diversity of these conferences will be greatly diminished.
To rectify the situation, conferences should be returned to university campuses where there are many classrooms that could be used as conference rooms. The professors and students could help organize and provide services for the conferences. Lodgings around universities typically are relatively inexpensive. It would be possible to reduce the budget for holding conferences and decrease the attendance fees.
The IEEE must do something to avoid making its conferences a rich man's club. Otherwise, please remove the "non" from the "About Us" statement as it seems very ironic these days.