Thousand+ core cluster of Raspberry Pi

Discussion in 'Hardware' started by nitro, Dec 12, 2015.

  1. nitro

    nitro

    For $110, this man built an earthquake warning system
    A professor at UC Berkeley (in California's earthquake country) has created a prototype device that warns of pending quake-related rumbles and could be installed as easily as a home fire alarm.

    "...Bloom, anastrophysics professor at UC Berkeley, cobbled together a Raspberry Pi single-board computer, an SD card, wired power speaker, and mini Wi-Fi adapter for his low-power earthquake-early-warning (EEW) device, which can run for up to days at a time on a USB battery. The gadget taps data from the ShakeAlert system, a prototype EEW for California now being beta-tested by the Bay Area Rapid Transit System, Google, and about 40 California scientists and engineers, including Bloom...."

    http://www.cnet.com/uk/news/for-110-this-man-built-an-earthquake-warning-system/
     
    #31     Sep 19, 2016
  2. I've just checked the performance of my system: 250 million bid/ask quotes per second. It's a Java program, running on a quad-core CPU. What in the world is taking so long that you can do only 10 quotes per second?
     
    #32     Sep 22, 2016
  3. Lol. No idea what you're measuring, or think you're measuring...
     
    #33     Sep 22, 2016
  4. Suppose you have a historical data set, which consists of tick prices. Let's say the size of this historical data set is 1 million entries (i.e. bid/ask prices). Now, let's say that with that historical data set, you want to optimize a simple MA-crossover strategy, which has 2 parameters: faster MA and longer MA. If you set the range of MA length to [1..100], that would give you:
    100 * 100 = 10,000 different trading strategies

    So, you want to backtest 10,000 strategies using a historical data set which has 1 million data points. Basically, this means that you need to go over each one of the 1 million data points 10,000 times, which gives you a total of 10 billion iterations.

    Since my system can process 250 million quotes per second, this backtesting job would take:
    10,000,000,000 / 250,000,000 = 40 seconds

    Nitro's system, on the other hand, would take:
    10,000,000,000 / 10 = 1,000,000 seconds = 11.6 days

    So, I was just wondering what Nitro is doing that takes so long.
     
    #34     Sep 23, 2016
  5. Zithiel89

    Zithiel89

    But what it really can do trading wise ? Is it some light Linux operated or how really ? I simply do not get it at all. So please do if you can all this little things. "With its Broadcom BCM2835 application processor (1GHz ARM11 core), 512MB of RAM, a microSD card slot, a mini-HDMI socket supporting 1080p (at 60 frames per second), micro-USB sockets and an identical pin layout to its larger Pi siblings, the Zero can do plenty of heavy lifting, despite its tiny size."
     
    #35     Jun 24, 2019
  6. 250mm quotes/second from where? The US markets in 2015 peak at around 15mm.
     
    #36     Jun 25, 2019