Thursday August 3rd

HP and the Machine

Hello…from Palo Alto!

  • Okay, sort of a lie. I went to Mountain View first, and got a mango slush from Teaspoon!

  • The Trip up was surprisingly drama-free. I tried a different service and it was quite nicer, with internet that worked. I think I’ll do that again next time.

Hung out a bit…

  • I went up to a Starbucks in Palo Alto and randomly met a CS professor from Stanford who is now retired, who is originally from the UK (British), named Caleb. He was quite kind, and after we met and shook hands, and I told him why I was in the area, he smiled and said to take care of myself, and continued working, tapping away on his laptop, as I worked on mine.

Then I went to HP, for a talk on their memory driven-computer, “The Machine”.

Advantages of Memory Driven Computing

  • Fast, persistent memory
  • Their system uses a photonic engine that eliminates distance as a hindrance (as opposed to using copper, where energy is dependent on distance) and creates otherwise impossible topologies
  • it’s task specific processing

Using Photonics

  • This copper stack is replaced by this photonic cable (the thin one above it)

The photonic engine (with scaled-up version)

Photonic engine (with my fingers to scale)

Data Structures..how do we program such a system

  • We use pointers because we can directly access memory. There is no need for serializing and deserializing data. For example, in Java, no usage of “new”.

  • Susan also said that clearing the cache or turning off will not solve a problem as it would today, because of the persistence of memory. That’s an interesting problem / challenge.

Example using Managed (garbage collection) vs unmanaged (C++).

  • Uses data structure called MPGC (GC is garbage collection)

Java without serializing and deserializing data

  • Susan Spence mentioned that to share the data, you’d then give it a name and it would be available to other applications, because they share memory.

The idea of containers booting up within a nanosecond

  • She spoke about have a system in which shifting from one container to another is a process swap, so would take a nano-second instead of a minute. Perhaps renting a system to someone who needs that sort of access to that storage for say, only an hour.

  • There would be no disk communication latency because you would be working directly with the memory. That makes sense for numerical computation.

Data structures using Graph Theory

  • One of the concepts she spoke about very much reminded me of diffs or deltas in Github. So I chatted with her afterwards and she said they were very much looking into graph theory. I asked her why she thought there is a lot of a concentration in general in that area for solving computational problems, and she was saying because of the sort of complexities you get because of the intersection of attributes.

  • She spoke about using LSGI and other such algorithms. Also, SPARKLE, which uses a bit of Scala and Akka.

More pics (hardware)

The big boxes are FPGAs, which were used for protoyping…

  • They will be replaced with ASIC

Questions

  • If everything is in one memory centre, with such a large data size, what sort of time and what sort of checks will be used for vulnerabilities/ someone hacking into your system?

  • I hadn’t heard of neuromorphic computing before. Would like to find out more about that.

  • Want to know more about Fabric Attached Memory

Overall, a wonderful night!

  • A little bit of hype, but some incredibly smart people. I’m happy I attended, and I even saw Ken, my Xoogler friend! I also met some really nice robotics people.

The Lobby

The View

Katas

  • replace format thingy
def task(w,n,c):
    d = n * c
    if w == 'Monday':
        return 'It is {} today, James, you have to work, 
        you must spray {} trees and you need {} dollars 
        to buy liquid'.format(w, n, d)
    elif w == 'Tuesday':
        return 'It is {} today, John, you have to work,
        you must spray {} trees and you need {} dollars 
        to buy liquid'.format(w, n, d)
    elif w == 'Wednesday':
        return 'It is {} today, Robert, you have to work,
        you must spray {} trees and you need {} dollars
        to buy liquid'.format(w, n, d)
    elif w == 'Thursday':
        return 'It is {} today, Michael, you have to work, 
        you must spray {} trees and you need {} dollars 
        to buy liquid'.format(w, n, d)
    elif w == 'Friday':
        return 'It is {} today, William, you have to work,
        you must spray {} trees and you need {} dollars
        to buy liquid'.format(w, n, d)
  • Spongebob alternate case one
function spongeMeme(sentence) {
  var arr = []
  // push lower and upper case correction to arr
  for (var i = 0; i < sentence.length; i++)
  {
    if (i % 2 == 0)
    {
      arr.push(sentence[i].toUpperCase()) 
    }
    else{
      arr.push(sentence[i].toLowerCase())
    }  
  }
  // join arr to string
  var pete = arr.join('')
  return(pete)   
}
  • return highest num in arr and throw err if arr.length is 0
function highestNumb(arr) {
  if (arr.length == 0){
    return one;
  }
  arr = arr.sort(function (a, b) {  return a - b;  });
  var a = arr.length - 1
  if ((arr.length == 0) || (arr == [])){
    return 1/0
    }
  else{
  return arr[a]
  }
}
  • hex to int
def hexa_first_principles(hexadecimal):
    a = int(hexadecimal, 16)
    return a
Written on August 3, 2017