Answers by Paul Lu for the CISS Team
|
We're thrilled that you are interested in our experiment! This FAQ is meant to quickly answer some of the most common questions that have come up on Slashdot and elsewhere. As you can imagine, we are busy preparing for November 4, so our answers are brief, non-refereed, and not suitable for quoting in an academic venue (not even as “personal communication”). If you are interested in a quote on a specific issue, please contact me directly at paullu@cs.ualberta.ca. (I will try, but I may fail, to answer promptly. Sorry.) |
Selected Media and Other Links:
The Team:
Aaron Davidson, Mark Goldenberg, Aiko Huckauf, Wolfgang Jäger, Rob Lake, Paul Lu, Chris Pinchak, Jonathan Schaeffer, Yunjie Xu, Paul Masiar and C3.ca, and many kind, helpful systems administrators across Canada.
Also, thanks to Nolan Bard, Jeremy Handcock, Mark Lee, George Ma, Danny Ngo, Yaling Pei, Jeff Siegel, and Victor Salamon.
This list may be incomplete. Our apologies for any unintentional omissions.
FAQ:
What is CISS?
CISS is the working name of the project and it stands for Canadian Internetworked Scientific Supercomputer. It is pronounced the same way as “kiss”. We reserve the right to change the name or meaning as soon as we think of something better.
CISS is an attempt to build the software and social infrastructure for a Canada-wide metacomputer. We do not have any current plans to go beyond Canada.
The (eventually, open source) software for CISS is developed as part of the academic Trellis Project in the Department of Computing Science, University of Alberta.
What is a metacomputer?
Put simply, a metacomputer is an useful aggregation of individual computer systems. The CISS metacomputer aggregates high-performance systems across Canada to perform a scientific computation.
What is CISS-1? What happens on November 4, 2002?
On November 4, 2002, the scientific computation running across Canada will be an application in computational chemistry. I will refer to this first experiment as “CISS-1”. Future CISS experiments will feature different applications.
We have already run smaller experiments across the provinces of Alberta and British Columbia, but CISS-1 is when we go nation-wide.
How new is CISS and metacomputing?
As a project, CISS is fairly new.
In computing science, the dream of metacomputing has been around for decades. In various forms (and with important distinctions), it has also been known as “distributed computing”, “batch scheduling”, “cycle stealing”, and (most recently) “grid computing”. Some well-known, contemporary examples in this area include SETI@home, Project RC5/distributed.net, and the projects associated with Globus/Open Grid Service Architecture (OGSA).
Of course, there are many, many other related projects, including some in Canada (for example, Grid Canada). Google the above terms or look on Slashdot to find some of them.
How is CISS different from SETI@home and Project RC5?
SETI@home, Project RC5, and similar projects are based on single applications (e.g., signal processing, cracking codes).
CISS-1 will also have a single application. But, our software and social infrastructure is designed to support any application. In future CISS experiments, we will support different applications, possibly running as part of the same experiment (e.g., computational chemistry and physics “at the same time”).
Why do you not refer to CISS as “grid computing”?
Frankly, what we are trying to do is considerably more modest and simpler than some of the definitions of “grid computing”. Therefore, I prefer to use the older terminology (i.e., metacomputing) to reflect our more limited scope. I might informally use the term “grid computing” when discussing CISS, but that is because “grid” is the more familiar term these days.
Currently, we do not use any of the new software that might be considered part of “grid computing”. However, the design of CISS allows us to incorporate and/or co-exist with grid technology, in the future.
So what makes CISS unique?
The academic, scholarly answer to this is complicated. You can read one of our academic papers for a partial answer.
The geek answer is this: We have tried to write the minimal amount of new software in order to implement CISS. This design philosophy is not due to laziness; we have carefully made our design decisions to make this possible. We have tried to build CISS on widely-deployed, existing software systems so that systems administrators who want to participate in CISS do not have to install new software. In fact, all that we have asked for is an account on their system. We don't require root access. That's it. (OK, we've asked for bash and ssh to be available, but we haven't required AFS, LDAP, or package X, Y, and Z.)
For example, we use Secure Shell, scripts, and a batch scheduler, if it is available. If there is no batch scheduler, we can deal with that too. Basically, almost any Unix system and any Linux box with a standard distribution is ready to go, as-is; just give us an account. The above assumes that the hardware has enough resources for the application itself.
In our experience, when dealing with a diverse collection of people, administrators, and systems, the fewer the requirements for participating, the easier it is for people to agree to participate. (See topic “social infrastructure” below.)
Who developed the software for CISS? How much development has gone into CISS?
The CISS software is developed as part of the academic Trellis Project in the Department of Computing Science, University of Alberta.
Despite the minimalism of our software, a considerable amount of software development has been done for CISS-1. Chris Pinchak did most of the implementation, with substantial help from Mark Goldenberg, myself, and others. Overall, we are a small group of developers/researchers.
Of course, a considerable effort has gone into the formulation and preparation of the computational chemistry application too. For example, we are using the MOLPRO package (see link at top) and Dr. Wolfgang Jäger's group has spent a considerable amount of time towards making CISS-1 a success.
What do you mean by “social infrastructure”?
The software development has been substantial, but much more time has been spent to convince people to include their systems in the CISS experiment, logistics, arm twisting, etc. Paul Masiar, C3.ca, and Jonathan Schaeffer did a lot of this work. We are hoping that, if CISS-1 succeeds, then CISS-2 will be easier from the “people” point of view.
If our software was more complicated or demanding of the systems administrators, we believe that this logistical overhead would be even greater.
We accept, and are trying to work with, human nature. Technologists ignore human factors at their own peril.
“Why didn't they just make a client program for distributed computing so the entire country/world could help out?” (From a Slashdot posting.)
First, we had to keep CISS-1 simple enough for us to manage. Second, the computational chemistry application has significant resource requirements (e.g., large memory, significant disk space, etc.). Third, we are not interested in “cycle stealing” for CISS-1; the machines that we use will be dedicated to the task at hand.
How is CISS different from cluster computing or a Beowulf?
CISS works at one level of abstraction above clusters or Beowulfs. We can (and want to) aggregate individual clusters and Beowulfs into a single metacomputer. Also, CISS is designed to work across different administrative domains, whereas most (but not all) clusters are operated by a single systems administrator and are located within the same building.
It's not a question, but someone said, “Based on the article I would assume that they have made a custom tailored system (if not kludge) for one specific purpose.” (From a Slashdot posting.)
A reasonable comment, given what was said in the newspaper article. We're grateful for the interest of the Edmonton Journal!
But, to answer the implied technical questions: Some of the software for CISS is custom, but it is not limited to a single application. Yes, CISS-1 will only use a single application, but that's a policy issue instead of a mechanism issue. Is it a kludge? One can decide for oneself when we (eventually) release the software, write more documents, etc.
What are the communication properties of the application for CISS-1?
The computational chemistry application uses MOLPRO. Our thanks to the developers of MOLPRO for their support!
For CISS-1, the problem has been partitioned such that is it “embarassingly parallel”; there is no communication between different processes running MOLPRO, except to start up and finish the job. This is capacity computing; it is not capability computing. This is a simplifying assumption so that we can focus on the issues most relevant to CISS and the Trellis Project. But, as I say to my students, one should not be embarrassed if one can formulate the problem in an embarassingly parallel way; one should be happy that time was not wasted in parallelizing it in some other way.
Future CISS experiments may feature applications that are not embarassingly parallel. As far as the CISS/Trellis software is concerned, that's mostly (but not entirely) an application issue.
Someone quipped, “UofA called me and asked me if I still had my Commodore Amiga and could they borrow it! ;-)” (From a Slashdot posting.)
Actually... :-)
We are looking for someone in Inuvik or Iqaluit or even on some research ship in the Canadian Arctic who wants in on CISS-1. An ISP? Someone? Almost any Unix box will do, but the easiest thing for us (at this time) is a Linux box with at least 256 MB of RAM, 3 to 4 GB of free disk space, and a basic Internet connection. Please email me at paullu@cs.ualberta.ca.
Anybody at the South Pole? The International Space Station? Surely, there's a Linux box with a simple network connection!
“Why not join up with other universities that have been pursuing similar projects and give Canada access to the computing power of other countries as well?” (From a Slashdot posting.)
An excellent and fair question.
In science and engineering, some diversity of approaches is good. The Trellis Project is a research project. I doubt anyone (especially not us) will claim that all of the problems have been solved. Therefore, we are trying to find new and different ways to solve problems.
We have a great deal of respect for the many projects that came before us, including Condor, Legion, Globus, PUNCH, UNICORE, and many, many others. And, we will happily co-exist with other systems. For example, I am told that a Globus/grid-enabled Secure Shell is now available. Using that, the Trellis software could (in theory, I have not actually tried it) create a metacomputer out of a grid and also include computers that do not run any grid software.
In terms of software, we want to live up to one of our catch-phrases: “You want to be part of the metacomputer, then just give us an account; we don't need any special software.” There is still plenty of work to be done to fully live up to this goal, but it is still a goal.
What kind of computational chemistry are you doing?
From our chemistry collaborators:
Computational determination of chiral molecule interaction energies. An object that cannot be superimposed with its mirror image is called chiral. Our hands are the most familiar example of such chiral objects. Image and mirror image are called enantiomers of the chiral object. Chiral molecules play important roles in nature. For example, life itself is based on single enantiomeric forms of amino acids and sugars, a fact known as “homochirality of life”. Many pharmaceuticals are chiral molecules of which one enantiomer has a beneficial effect whereas the other may have severe side effects.
At the heart of it all is chiral recognition or chiral discrimination. A handshake is possible between two left or two right hands, but not between left and right hand. On the molecular level, there is an energy difference between the interaction of, for example, the right-handed form of a chiral molecule and the two enantiomeric forms of another substance.
How can this subtle energy difference have such profound effect in nature? The CISS-1 experiment gives us an opportunity to shed light on the answer. We will compute the interaction energy of two chiral molecules at many different separations and orientations. The specific system will consist of an aziridine derivative and hydrogenperoxide. The calculations will be carried out on two different enantiomers of the aziridine derivative. An analysis of the results will give a “chiral recognition surface” that identifies the active sites of the chiral components.
The chiral recognition surface will be determined by calculating the interaction energy at many different orientations and separations of the two molecules. Because of the complexity of the molecules, the energy needs to be calculated at many thousand points. Each point may require several hours of computing time on a modern workstation. In computational terms, this is an ideal parallel problem which can be distributed onto as many processors as points needed and is therefore ideally suited for the CISS experiment.
The efficient MOLPRO code for electronic structure calculations will be used in the experiment. Each single point calculation is exceedingly complex. The computations will be performed on a state-of-the-art level of theory.
We thank the authors of MOLPRO, P. Knowles and H. Werner, for permission to use their program suite for this experiment.