Grid Computing, week 3, saw another jam-packed lecture that left me with a great deal of questions. I think this is a good thing as it forces students to do required readings or simply be left behind in the lectures. Unfortunately I have not been keeping up with readings on this subject and am feeling the pressure now.

The lecture focused on a possible real life example of a butterfly grid. When developing such a grid, the key considerations must be identified:

  • Latency
  • Downtime
  • Security (cheating)
  • Physical scalability
  • Specific stakeholder needs (Production company, vendors, ISP, gamers, etc)

Soruce: http://pmlab.iecs.fcu.edu.tw/~cyt/cyt/data/GridApplication.ppt

The butterfly grid add 7 major tiers in the architecture… There was no diagram in the lecture to assist in understanding where these layers fit, a simplified representation can be seen above.

The layers listed in the lecture slide were:

  • Object Management System [OMS]
  • Network Protocol Stack (UPD/IP instead of TCP/IP for reduced latency)
  • Gateway Servers
  • Daemon Controllers
  • Game Servers
  • Data Store and Grid Service – existing layers
  • Globus Toolkit Services

Some news article relating to butterfly grids from early 2000s:

http://www.gridcomputingplanet.com/news/article.php/1107421/IBM-Butterflynet-Announce-Video-Game-Grid.htm

https://www-304.ibm.com/easyaccess/ukcomms/gclcontent/gcl_xmlid/91353

There is however very little recentĀ  information on Butterfly grids to be found on the web.

Even with second review of the lecture, the explanation of theĀ  butterfly grid is difficult to understand. Additionally the is very limited information on the web.

Source: Week 3 Lecture Slides

After discussing the butterfly grid in some detail, we took a look at a different application for grid computing (one I think more apt), Protein Crystallography. This part of the lecture was much more understandable.

The tutorial involved working in groups to understand some methods and challenges for job allocation amongst grid nodes. Even with the use of interface with Globus toolkit resource utilities I imagine the efficient allocation of resources is very hard to achieve and can see why the more dynamic approach of cloud computing is becoming more popular for most applications.