Monday's Notes

Monday was the first day of lectures. For the most part, a general introduction to the summer school, the concepts behind grid computing, and the large project on the last day of the school. My favorite presenter was Miron Livny, though all were very interesting. I took notes during the presentations, and I'll try to summarize the main points below. As a disclaimer, just because I took notes does not mean that I fully understand what was discussed. If anyone sees any errors or would like to discuss something, please feel free in the comments or if you see me around the school.

Introduction - Per Oster:

This presentation was a brief overview about European grid efforts and their progress and general structure. He explained the grid "ecosystem" that exists in Europe, where the different pieces that make up the grid community is organized into pyramid-like layers organized by the amount of use (with personal and local grids on the lowest and largest layers, and national grids on the highest/smallest portion).

He also spoke about the European Grid Initiative (EGI) that is trying to unite Europe's many different grids (Omii and Egee are just two of the many large ones). They have decided to start their effort under Dutch law, and they're having a meeting sometime this week to discuss the plan.

Distributed System Principles - Miron Livny:

I said earlier that Dr. Livny was my favorite presenter, and I hope that I can make my reasons clear in this short description.

He started out with a smart remark about how the US had already figured out what Europe was hard at work on (regarding the EGI). Dr. Oster didn't look too happy after hearing the comment, but everyone realized that it was...mostly...a joke.

His main point was that although there have been great technological advancements over the years, the principles behind their development and the general problems that they are designed to solve have not really changed. He talked a bit about the cycle that computer science has followed since its development: single user terminals to batch jobs to PCs with a single processor...but now with grids and PCs with multiple cores, we are back to batch jobs.

There was an interesting sentence on one of his slides:

Re-labeling, buzz words, or hype do not deliver as the problems we are facing are fundamental and silver bullet technologies are pipe dreams.

He discussed some of the main problems with grid computing. Problems as in...things that should be considered when designing a grid, or things that having a grid is supposed to solve, or things that might be made more difficult because of the grid. These were: data movement, identity management, and debugging.

He also went through the "benefits for distributed systems" outlined by a paper written by Enslow in 1978 (and pointed out how we are still having trouble with fully realizing them). He also discussed a list of the requirements that a grid must meet to actually be a grid: multiplicity of resources, component interaction (loose coupling), unity of control (desire to achieve a common goal), system transparency (the appearance of a single virtual machine to the user), and component autonomy (ability to refuse jobs based on its own policies). To tie these two points together, he said to be wary of a grid system that claims to have all of the benefits that Enslow proposed, because it probably doesn't meet the requirements.

Then there was just some random stuff in his presentation that I thought was interesting. Dr. Livny said that he is more concerned with High Throughput Computing (HTC) than High Performance Computing (HPC) because researchers usually express their needs for the system by asking "How many simulations can I run in X amount of time", instead of "How fast can I run this particular piece of code". So, optimizing your grid to provide good throughput should not be done as an afterthought, and that FLOPY != (60*60*24*7*52) * FLOPS.

Anyway, very interesting talk.

Distributed System Architectures and Scenarios - Miron Livny:

This talk was more specifically about Condor. Mentioned that one researcher solved the NUG30 problem in 6 days, 22 hours using Condor...a task that amounted to 11 years of work.

Then he went through the general steps for a Condor submission:

• user submits a job to schedd


• startd assigns a resource to schedd


• shadow is delegated a job and resource


• shadow sends both to the starter

Some new information...new to me at least...was that if a workflow engine is submitted, it can submit jobs. Also, that the grid manager can submit jobs to Globus or another Condor pool.

t-Infrastructure and Security(Practical) - Emidio Giorgio:

This discussion was mostly getting us set up with our user names and explaining about how to request an X.509 certificate. To request one, first contact the Registration Authority for your country become authorized, and then contact the certificate authority to actually receive the certificate. He also discussed the main security issues involving grids: DDOS attacks, illegal data distribution, and viruses.

General Introduction to Technologies - Per Oster

One thing that had nothing to do with the presentation that I found interesting was a word on one of the slides. I don't remember if it was the town that Dr. Oster was from, or what. The word is Otaniemi. I had a professor at Clemson named Dr. Hedetniemi, and on the first day of class he said that Hedet meant...something in Finnish, and that Niemi meant...river, I think. So the name of the town must mean something-river. Not a very good story since I don't remember the details, but it was interesting to me and I thought I would share.

Anyway, this discussion was an outline of the different technologies that we are going to cover during the school. Here were the important connections that he made:

• UNICORE - job submission and execution management


• Condor - HTC


• Globus - Service-oriented architecture


• gLite - Distributed Data Management


• Nordugrid/ARC - HPC


• OGSA-DAI/SAGA - Higher level APIs and metadata management


• Workflows - organizing work, P-GRADE portal

I'm not entirely sure of how much sense that makes to outside readers, but it's exactly what I have written in my notes. I'll be able to explain more in the coming days.

Introduction to Integrating Practicals - David Fergusson

The last event of the day was a short discussion of the practical, the final project at the end of the school. It was a summary, and doesn't make very much sense at the moment because it was just supposed to give us some idea so we can think about it over the next two weeks. Here is my understanding of the problem. we are going to be given a large coordinate plane split into sections (one for each technology that we discuss) with lots of background noise. Our job is to find pillars, rectangular objects hidden in the noise. On these pillars are blocks, and on these blocks are RDF triples. These triples can be used to search through an ontology to find text. The text is our answer. I don't know how many pillars there are, but our job is to find all of them. Reminds me of some sort of spy mission.

-----------------------

OKAY. That was a long recap. I hope it wasn't very boring...maybe these whole day summaries aren't such a good idea. We also went to Antibes, but Cole did a nice job of summarizing that, so I won't say any more about it. I took some pictures too, but mine were also dark. I didn't want to be the "stupid American" taking pictures of everything. No offense, Cole. :)