Updates from cbrinton RSS Toggle Comment Threads | Keyboard Shortcuts

  • cbrinton 4:17 am on October 21, 2012 Permalink | Reply
    Tags:   

    “Viralizing” a Youtube video 

    A few weeks ago, I attempted to increase the popularity of my Youtube video of myself playing guitar (http://www.youtube.com/watch?v=q5aA7A-ezZM&feature=plcp). My methods for doing so were:

    1. Advertise the link on my own facebook.
    2. Advertise the link on the network20q facebook.

    I did (1) first, and once the rate of view increase dropped down, I seeded (2) a few days later.

    The effect from my own facebook was not very pronounced. I got around 20 more views, a few likes, and one comment. The reason for this is probably that I had already advertised my video from my own facebook back when I posted the video to begin with, back in March of this year.

    On the other hand, seeding from the network20q account saw more profound results (thanks to all of you!). In two weeks, I was able to virtually double my view count, which increased from 150 to 285.

    I think this emphasizes the role that the various centrality measures play in influencing behavior: The network20q account beats my own quite drastically in terms of node, eigenvector, closeness, and betweenness centrality. As a result, seeding the link there is more likely to accrue a larger audience. Additionally, the educational “influence” power of the network20q account is much stronger than my own.

    VN:F [1.9.22_1171]
    Rating: 5.0/5 (1 vote cast)
     
    • Holly 4:04 pm on November 12, 2012 Permalink | Reply

      Not to mention, not everyone is on FB.

      VA:F [1.9.22_1171]
      Rating: 0.0/5 (0 votes cast)
  • cbrinton 10:39 pm on April 3, 2012 Permalink | Reply  

    Solving the wavelength assignment problem with a minimum number of wavelengths 

    Over the weekend, we were presented with the problem of wavelength assignment to end-to-end lightpaths in an optical network for the internet backbone. Obviously, each lightpath must have one wavelength, and in addition, no two lightpaths that share a link can have the same wavelength, since this is a Wavelength Division Multiplexed (WDM) scheme.

    Part of this had us recognize that we could represent the wavelength assignment problem as one of graph coloring using a transformed graph, where the nodes became lightpaths, and edges existed between two nodes a, b if a and b shared a link. In this way, no two neighboring lightpaths could have the same color.

    A question still remains: What is the minimum number of wavelengths (i.e. colors) we can use to solve this problem?

    Today in ORF 523 we investigated this problem. It is NP-hard. The most efficient and tractable way known to solve this problem is to break the graph down into its stable sets (subsets of the nodes such that no two nodes in the subset are connected), and solve the following optimization problem:

    minimize(sum of color variables, where one color variable is given to each stable set)

    subject to(each vertex must get assigned one color; 0 <= color variable <= 1; color variables are integers).

    So, why is this problem NP-hard? At a first glance, it looks like an LP. The problem is that each of the variables must be integer - More specifically, 0 or 1. This is a decision problem – You cannot have a fraction of a color. A stable set is either assigned a unique color, or it is not.

    This leads us into another branch of optimization called integer programming (IP, or MIP for mixed integer / linear programming), which constrains the feasible region of the optimization problem to all integer points within the polyhedron.

    To obtain a solution to this problem, most computer codes will solve the LP-relaxation to this problem (i.e. remove the integer constraint) by applying the procedure of column generation, whereby the dual LP is taken and solved repeatedly until a feasible solution to the primal is obtained. Once the solution to the LP-relaxation is obtained, the code will apply the technique of branch and bound, where at each stage another LP is solved by restricting the branch variable to be 0 and 1, respectively.

    More information on the graph coloring problem and branch and bound, refer to http://www.optimization-online.org/DB_FILE/2010/03/2568.pdf.

    VN:F [1.9.22_1171]
    Rating: 4.0/5 (4 votes cast)
     
  • cbrinton 8:58 pm on March 24, 2012 Permalink | Reply  

    Energy Network Optimization at CISS 

    Friday afternoon at CISS had an interesting session on energy network optimization. One talk I found particularly interesting was Steven Low’s discussion on the Optimal Power Flow (OPF) problem.

    The OPF problem is a non-convex problem that has been widely explored by both power systems and optimization engineers since the 1960s. It deals with finding an optimal operating point of a power system, in terms of non-linear, physical parameters (such as active power, reactive power, as well as voltage and current magnitude and phase) to minimize an objective function such as generation or transmission cost. We can see an analogy here with utility optimization in TCP rate allocation studied in Chapter 14, where we are maximizing network utility subject to flow capacities to find the optimal rate allocation to each session (i.e. the ‘operating point’ of the network).

    Steven Low and his group at Caltech have successfully exploited the physical properties of power systems to obtain a polynomial time algorithm to find a global optimum to the OPF problem for a large class of power systems. They propose a semidefinite programming (SDP) optimization, which is the dual problem of an equivalent formulation of the OPF. We know from class that the Lagrange dual problem is always convex, regardless of the structure of the primal problem. However, one must still concern themselves with the “size” of the duality gap.

    A major breakthrough in stated Steven Low’s discussion was that his group found a necessary and sufficient condition for the duality gap to be zero, meaning that a global optimum of the primal can be retrieved from a solution to the dual. Interestingly, this condition is met for five IEEE test bus systems (varying from 14 to 300 buses in size) tested in their paper.

    To grasp this condition in a nutshell, we must consider four optimization problems they develop in their paper:

    Optimization 1 (non-convex): A non-convex equivalence of OPF.

    Optimization 2 (non-convex): A change of variables from the vector of bus voltages to a matrix W = X*Xt, where Xt is the transpose of X.

    Optimization 3 (convex): A convex relaxation of Optimization 2, which removes the constraint of rank(W) = 1.

    Optimization 4 (convex): The dual of optimization 3.

    Now, the crucial point: They prove that the duality gap is zero if the solution to Optimization 4 results in a positive semi-definite matrix with eigenvalue 0 having multiplicity 2. As long as this is satisfied, we can backtrack and retrieve a global solution to the primal problem. I found this to be particularly interesting – If they can prove that this will work for any power system, then they will have found a polynomial-time algorithm for solving the OPF, which would be of great utility to both network and power systems engineering.

    A link to their paper can be found here: https://www.cds.caltech.edu/~lavaei/TPS_1.pdf

    VN:F [1.9.22_1171]
    Rating: 5.0/5 (4 votes cast)
     
  • cbrinton 8:32 pm on March 10, 2012 Permalink | Reply  

    Paris Metro Pricing, The Internet, and First Class Flying 

    The other day I was looking into the possibility of purchasing first class tickets FAR in advance for a potential trip to California, thinking that maybe the prices would be less. My intuition was soon proved wrong, as I found that a flight from Newark, NJ to San Francisco, CA first class for June would cost ~$1,600 per person (round trip), as opposed to ~$500 per person (round trip) for economy seating – First class is over three times the price!

    We can explain this phenomenon in terms of Paris Metro Pricing: Logically, for any given customer, the demand for a seat will decrease as the price increases. Therefore, there will be less people willing to purchase the first class tickets, so a first class cabin will be less crowded.

    However, there are other notions here that are not directly explained through the basic Paris Metro Pricing model. For instance, the first class section of the flight enjoys a lot more luxuries than the rest of the plane. For one, the seats are bigger. Additionally, certain meals, bathrooms, and even movie services are limited to only first class fliers. In terms of economics, we can roughly see this as a vertical and rightward shift on the demand curve – As more services and conveniences are offered to this cabin, the demand will increase for a fixed price.

    There’s a nice quick read that I found on Paris Metro Pricing and the internet: Paris Metro Pricing. The basic idea is that we add multiple links, and charge users more for using certain links to obtain higher quality of service. However, we can’t add too many extra links, because then some of the advantages reaped by using statistical multiplexing on a packet-switched network would not be observed. An underlying assumption is that “consumers…[must be] willing to accept occasional large deviations from the expected quality of service”.  Is this a valid assumption to make? The author justifies it using two different arguments:

    1. In other situations, users who pay more for higher QoS tend to accept this occasional deviation. For instance, people still buy first class airline tickets even with the potential of a baby crying on board somewhere.

    2. Fully guaranteed QoS at all times is unrealistic on the Internet. There’s never a 100% guarantee that data will receive its destination, as a separate link is never opened for a given connection, as in circuit-switched networks.

    But, is this really a justified assumption? If I was paying even an extra $5 – 10 dollars a month for my internet, I would expect to have extremely high QoS at all times. ISPs would have to come up with a scheme whereby they would reimburse a certain amount of this money given some scalar which quantifies the ratio of the observed to expected QoS in a given month. For instance, in simple case,

    Amount charged/Month = (Base Price) + (Incremental QoS Price)((Number of successful QoS responses to requests)/(Total number of requests))

    In a more complicated scheme, we would further break down “successful” into what percentage of “successful” we had actually obtained in each case.

    VN:F [1.9.22_1171]
    Rating: 5.0/5 (1 vote cast)
     
  • cbrinton 5:35 pm on March 3, 2012 Permalink | Reply  

    SES launches their 50th Geostationary Satellite into Orbit 

    Satellite Services (SES) is a leading satellite operator, “providing reliable and secure satellite communications solutions to broadcast, telecom, corporate and government customers worldwide”. Headquartered in Luxembourg, they own 50 geostationary satellites that are monitored and controlled by their offices located around the world. In the US, they have two offices – one in Washington DC, and one right here in Princeton.

    On February 15, SES launched their latest satellite, SES-4, into orbit from Baikonur, Kazakhstan. SES-4 became the company’s 50th satellite in their fleet. It is their largest to date, and will replace the current NSS-7 satellite at 338°E. With a life expectancy of over 15 years, it is specifically designed to “enhance SES’ ability to provide fixed satellite services to…the Americas, Europe, Africa, and the Middle East”. It is equipped with transponders to operate in both the C-band (4 – 8 GHz) and the Ku-band (11 – 15 GHz). Typically, geostationary satellites operate in or near the Ku-band. It is interesting to note that operating in the C-band is better under adverse weather conditions than the Ku-band.

    The video of the launch can be seen at http://www.ilslaunch.com/newsroom/video-gallery/ses-4-launch-highlights. To learn more about the company and their satellite coverage, you can visit their website / blog at: http://www.ses.com/.

    VN:F [1.9.22_1171]
    Rating: 3.0/5 (2 votes cast)
     
    • Jean Steger 11:02 am on December 31, 2012 Permalink | Reply

      Things i have seen in terms of personal computer memory is the fact there are features such as SDRAM, DDR and the like, that must fit in with the specific features of the mother board. If the pc’s motherboard is very current while there are no computer OS issues, updating the memory space literally requires under an hour or so. It’s among the easiest pc upgrade techniques one can imagine. Thanks for discussing your ideas.

      VA:F [1.9.22_1171]
      Rating: 0.0/5 (0 votes cast)
  • cbrinton 5:23 am on February 26, 2012 Permalink | Reply  

    A Remarkable Effort to Define the term “Smart Grid” 

    As I sit in my living room writing this blog, I look around at my personal items in the room: three electric guitars, two desktop computers, a laptop, my iPhone, lights, etc. What do all these devices have in common? They require power, in one form or another, to operate.

    Our society has become heavily dependent on both power and communications. But even the “fuel” of communications is electricity. Indeed, without power, our society could not exist as we know it.

    One common theme between power systems and communications is that the current infrastructure in many instances limits the development of future technologies. Specifically for the former, one must justify the added cost of replacing existing transmission lines or generators in order to get this accomplished. Alternatively, one must find ways of incorporating new ideas into the existing grid, which presents challenges of its own. For example, it is not an easy task to build a localized generator and interface it with the existing grid.

    But the problems that the current electric grid faces are far more reaching and diverse than the mere fact that its “old”. The most dominant form of electricity generation in the US by a long shot is coal, and not far behind it is nuclear energy. Why is it that the two most harmful forms of energy – coal because it burns fossil fuels, and nuclear because of the hazards it poses – are the most common forms of generation? Because they are also BY FAR the cheapest, and in addition, our current infrastructure has many of these forms of generators already in place.

    As more and more devices run on electricity, and with the world’s population at unprecedented levels, you may wonder what effect this could have on electricity distribution. By some estimates, the electricity demand is expected to increase by up to 1% per year over the next 20 years. As a result, we will need more generators and more ways of providing the added electricity reliably to prevent blackouts. Speaking of reliability, what is one way to reduce the time it takes to determine the root cause location of a blackout? Incorporating “smarter” sensor devices into the grid in key locations, so that once a line, transformer, or other device fails, the power companies can immediately determine where the fault has occurred.

    Fuel prices are expected to rise to unprecedented levels in the near future – Up to $4.00 a gallon for regular. Is there any way to obviate the need for gasoline in cars? One plausible alternative is to operate cars solely or partially (hybrid) on electricity. But this also poses various problems, from the fact that electricity itself is expensive to the fact that eliminating the fuel economy in the US could turn the nation upside down.

    The MIT Energy Initiative provides a comprehensive portrait of the US Electricity Grid, its current state, as well as the challenges it is likely to face over the next decades. In doing so, it defines various aspects of what we have coined “Smart Grid” in everyday speech (for engineers, at least).  Their current issue (one of six) describes the following:

    1. The potential of integrating renewable sources of energy (such as wind and solar power) into the grid.

    2. Proposed ways of dealing with cybersecurity threats.

    3. Ways of improving grid efficiency through advanced metering devices.

    4. The pros and cons of distributed (i.e. smaller and more localized) generators.

    5. Key areas (grid design planning, computational tools, etc) in which research is still needed looking into the near future.

    You can find a link to their hour long video as well as their 280 page report at http://web.mit.edu/mitei/news/videos/electric-grid-study-2011.html. I hope you find it as interesting as I did.

    VN:F [1.9.22_1171]
    Rating: 5.0/5 (1 vote cast)
     
  • cbrinton 2:44 am on February 18, 2012 Permalink | Reply  

    AT&T Continues Optical Systems Research 

    The rise of 3G and 4G wireless telecommunication and data services at almost the same time that the photonic “bubble burst” happened in the 1990s caused many companies to shift research towards wireless networks and put less emphasis on improving the efficiency, capacity, and other factors of optical systems.

    Despite popular belief that companies such as AT&T have taken a complete halt on photonics research, there is an Optical Systems Research division of AT&T right here in Middletown, NJ, that investigates potentials of long-haul, wavelength division multiplexed networks as well as extending the reach of conventional time division multiplexed networks using hybrid amplification techniques (thoughstretching the standard notion of a PASSIVE optical network).

    A summary of their current research interests can be found here: http://photonicssociety.org/newsletters/aug11/RH_ATnT.html.

    Here’s one nice passage from the article, which indicates the number of fiber miles AT&T owns, and that the typical bit rate on their systems are 40 Gb/s:

    “AT&T operates one of the world’s largest fiber optic networks, with more than 886,000 fiber route miles worldwide for access, metro, long-haul and trans-oceanic communications. AT&T’s global backbone network carries more than 23.7 petabytes of data traffic on an average business day. The continental-US backbone network includes more than 340,000 wavelength-miles of 40 Gb/s transport. The Optical Systems Research Department in Middletown NJ, consisting of 10 full-time researchers, explores optical-layer innovation pertinent to all segments of this extensive network, with an emphasis on forward-looking topics critical to AT&T’s mission.”

    VN:F [1.9.22_1171]
    Rating: 0.0/5 (0 votes cast)
     
  • cbrinton 10:51 pm on February 9, 2012 Permalink | Reply  

    The Electric Transmission System as a “Network” 

    Here’s an interesting paper I found while doing research on electric transmission system optimization:

    http://www.eia.gov/oiaf/analysispaper/transmiss.html

    It concerns inter-regional power power trading among four regions in the northeastern United States, and the potential effects of this on electricity costs and trade level variation by region. From a networking perspective, its interesting to note how their model of the transmission systems can be related to a communications network, and how the same optimization tools can be used in both cases (subject to different constraints, of course).

    VN:F [1.9.22_1171]
    Rating: 5.0/5 (2 votes cast)
     
    • Zac 9:34 pm on January 14, 2013 Permalink | Reply

      Great solution it looks wonderful

      VA:F [1.9.22_1171]
      Rating: 0.0/5 (0 votes cast)
c
compose new post
j
next post/next comment
k
previous post/previous comment
r
reply
e
edit
o
show/hide comments
t
go to top
l
go to login
h
show/hide help
shift + esc
cancel