GRID

1
GRID

• FarhadJavidi

NCSET
2
What is GRID?

• Whereas the Web is a service for sharing
  information over the Internet, the Grid is a
  service for sharing computer power and data
  storage capacity over the Internet. The Grid goes
  well beyond simple communication between
  computers, and aims ultimately to turn the global
  network of computers into one vast computational
  resource.
• That is the dream. But the reality is that today,
  the Grid is a "work in progress", with the
  underlying technology still in a prototype phase,
  and being developed by hundreds of researchers
  and software engineers around the world.
• The Grid is attracting a lot of interest because
  its future, even if still uncertain, is
  potentially revolutionary.
• So the interest comes not only from experts in
  computer science, but from scientists,
  businessmen, journalists and, presumably, you,
  the browser of this website!

3
The Dream

• Imagine a lot of computers, let's say several
  million. They are desktop PCs and workstations,
  mainframes and supercomputers, but also data
  vaults and instruments such as meteorological
  sensors and visualization devices.
• Imagine they are situated all over the world.
  Obviously, they belong to many different people
  (students, doctors, secretaries) and
  institutions (companies, universities,
  hospitals).
• So far you have imagined nothing new. This is
  pretty much what the world looks like today.
• Now imagine that you connect all of these
  computers to the Internet. Still not much new,
  most of them are probably connected already.
• Now imagine that you have a magic tool which
  makes all of them act as a single, huge and
  powerful computer. Wow! Now that really is
  different. This huge, sprawling mess of a
  computer is what some dreamers think "The Grid"
  will be.

4
What if this dream came true?

• Well, if you are a scientist, and you want to run
  a colleague's molecular simulation program, you
  would no longer need to install the program on
  your machine. Instead, you could just ask the
  Grid to run it remotely on your colleague's
  computer. Or if your colleague was busy, you
  could ask the Grid to copy the program to another
  computer, or set of computers, that were sitting
  idle somewhere on the other side of the planet,
  and run your program there. In fact, you wouldn't
  need to ask the Grid anything. It would find out
  for you the best place to run the program, and
  install it there.
• And if you needed to analyse a lot of data from
  different computers all over the Globe, you could
  ask the Grid to do this. Again, the Grid could
  find out where the most convenient source of the
  data is without you specifying anything, and do
  the analysis on the data wherever it is.
• And if you wanted to do this analysis
  interactively in collaboration with several
  colleagues around the world, the Grid would link
  your computers up so it felt like you were all on
  a local network. This would happen without you
  having to worry about lots of special passwords,
  the Grid could figure out who should be able to
  take part in this common activity. 

5
What types of apps can Grid be used for?

• A simple answer is that just about anything you
  do on a computer you could also do on the Grid.
  After all, the Grid is, in some sense just a very
  big computer.
• A more sophisticated answer is that the first
  big-time users of the Grid will probably be
  scientists with challenging applications that are
  simply too difficult to do on just one set of
  computers. After all, the Grid was originally
  conceived for these types of users, and is being
  built by them, too.

6
A Brief History

• Many of the basic ideas behind the Grid have been
  around in one form or other throughout the
  history of computing. For example, one of the
  "novel" ideas of the Grid is sharing computing
  power. Nowadays, where most people have more than
  enough computing power on their own PC, sharing
  is unnecessary for most purposes. But back in the
  sixties and seventies, sharing computer power was
  essential. At that time, computing was dominated
  by huge mainframe computers, which had to be
  shared by whole organizations.
• In 1965 the developers of an operating system
  called Multics (an ancestor of Unix, which in
  turn is an ancestor of Linux - a popular
  operating system today) presented a vision of
  "computing as an utility" - in many ways
  uncannily like the Grid vision today. Access to
  the computing resources was envisioned to be
  exactly like water, gas and electricity -
  something which the client connects to and pays
  for according to the amount of use. Ironically,
  "utility computing" is all the rage again these
  days, and used more or less as a synonym for the
  Grid by some people.
• So, yes, there is a certain amount of
  "reinventing the wheel" going on in developing
  the Grid. However, each time the wheel is
  reinvented, it is reinvented in a much more
  powerful form, because computer processors,
  memories and networks improve at an exponential
  rates which are associated with Moore's law.
• Because of the huge improvements of the
  underlying hardware (typically more than a factor
  of 100x every decade), it is fair to say that
  reinvented wheels are qualitatively different
  solutions, not just small improvements on their
  predecessor.

7
Breaking Moore's law

• One of the most misused concepts in the Grid
  community, and in computing in general, is
  "Moore's law". Originally, this was a statement
  made by Gordon Moore, one of the founders of
  Intel, about the number of transistors that could
  be squeezed on a silicon chip. In 1965, Moore
  noted in an article he wrote that this number was
  doubling once every year, and over time, this
  doubling rate has been revised down to once every
  18 months.
• Regardless of the fact that Moore's statement
  was limited to a very specific quantity - the
  number of transistors on a chip - it is now used
  for just about everything else in computing.
  "Computing power is doubling every 18 months,
  according to Moore's law" is one common misuse of
  Moore's observation.
• Worse, comparisons are made - even in learned
  journals - between different quantities that have
  nothing to do with Moore's law. A frequently
  heard comparison is that network performance is
  doubling every 9 months, and data storage density
  every 12 months, both "outperforming" Moore's
  law. The implication is that somehow, computer
  processors are not keeping up with data storage
  and network capacity. This pseudo-trend is even
  used by some as an argument in favour of the
  Grid. The argument is that you will need to share
  processing power because your local processor,
  which follow Moore's law, will become
  "underpowered" compared to the data storage and
  networking capacity available to you.

8
Breaking Moore's Law

• In fact, this ignores a number of trends which
  Moore's law does not take into account. For
  example, the clock cycle of processors increases
  along with the increase in the number of
  transistors per chip. So processor power, however
  you measure it, grows faster than Moore's law.
  And improvements in chip architecture and
  operating systems also make processors more
  powerful than just the sum of their transistors.
• In short, detailed comparisons of different
  growth rates with Moore's law are often
  misleading. It is best to see Moore's law as
  simply a metaphor for exponential growth in the
  performance of IT hardware - which no one can
  doubt is true. As a result of this exponential
  growth, with every year that passes, the Grid
  concept becomes more feasible, because
  distributed processors can be more tightly
  integrated, thanks to improvements in network
  speeds.
• Of course, individual stand-alone computers also
  become more powerful with every year that goes
  by, so the sorts of problems the Grid is
  particularly useful for will change with time.
  Still, there will always be some problems whose
  complexity, or dependence on distributed
  resources, puts them out of reach of a single
  computer, but which could be tackled with a Grid.
  So Moore's law does not threaten the Grid, it
  merely shifts the domain of application of the
  Grid to increasingly complex problems.

9
Questions?