Renting a Supercomputer
- Thread starter aphexcoil
- Start date
Quote from canyonman00:
"Thoroughbred Handicapping With Confidence" (first edition) was completed in '95. I completed the last revision in '98. Originally I had envisioned a book for sale but it proved so successful that I decided against marketing it. It then became the outline/guide for further strategy development.
I then decided to turn it into proprietary software and I sat down with the folks at IBM (software division) late in '98. When we couldn't agree on how to proceed, I struck out on my own.
I now run the actual handicapping software in two flavors, Windows and Linux. The Beowulf cluster does all the real data crunching, sorting and assemblage and the windows desktop/laptop's just provide a Windows based onsite/remote access capacity.
The software provides and allows straight wagering as well as determining when I need a Dutch wager for higher success probabilities. And I do successfully wager regularly on the Hong Kong circuit. But it's still fun sitting down with the local OTB regulars and doubters and causing a raised eyebrow at the luck/success from time to time.
This is so cool!
You definitely deserve every dollar you win with this system.
The markets with greatest degree of inefficiency are 1) the most illiquid and 2) the newest markets that have just begun trading.
As is axiomatic, as markets mature and gain liquidity they tend asymptotically toward pure efficiency. Therefore the S&P and the Bonds is probably not he best place to look for an edge. If you insist on looking there you are probably going to be looking in the wrong haystack for your needle.
If I were you I would look at arb possibilities between SSF and their equity counterparts. This trade is good for maybe 1.5- 2 years before all the cream is skimmed. In this day and age marketable inefficiencies usually don't last for long. But make no mistake, they are there you just have to look in the right place. And you have to be quick on the scene.
Dr. Zhivodka
I just surfed the web and found an article I would like to share with people interested in Java programming and parallel computing. It gives you the possibility to build your own cluster and you even don't need to switch to Linux, because you can run the system on every OS where Java can run on.
The author shows how to use pseudo remote threads to distribute tasks to different machines. This article even contains sample code!
http://www-106.ibm.com/developerworks/library/j-super.html
Also a good starting point to build you own cluster (not Java specific) could be the SCL cluster cookbook:
http://www.scl.ameslab.gov/Projects/ClusterCookbook/index.html
The author shows how to use pseudo remote threads to distribute tasks to different machines. This article even contains sample code!
http://www-106.ibm.com/developerworks/library/j-super.html
Also a good starting point to build you own cluster (not Java specific) could be the SCL cluster cookbook:
http://www.scl.ameslab.gov/Projects/ClusterCookbook/index.html
THE CRASH OF 2002?
Computer Scientists from around the globe are meeting in emergency session tonight in Paris to find a way to stop the Boy Wonder aphexcoil from getting access to a Super Computer.
aphexcoil, the developer of âJasperâ, an API designed to work with Interactive Brokerâs TWS order platform, is thought to have caused some nine system wide outages on the Globex Exchange, and a two day power outage in the city of Baltimore.
The Boy Wonder has compiled tick data going back to the year 1622, and needs to rent time (Perhaps 19 to 20 months.) on a Super Computer in order to test his latest trading theories. Prof. Richard Rundy of The University of Pennsylvaniaâs Wharton School, told The Daily Skirt that the worldâs computer experts fear that if aphexcoil gets within ten feet of any Super Computer, the entire planetâs network could crash.
Computer Scientists from around the globe are meeting in emergency session tonight in Paris to find a way to stop the Boy Wonder aphexcoil from getting access to a Super Computer.
aphexcoil, the developer of âJasperâ, an API designed to work with Interactive Brokerâs TWS order platform, is thought to have caused some nine system wide outages on the Globex Exchange, and a two day power outage in the city of Baltimore.
The Boy Wonder has compiled tick data going back to the year 1622, and needs to rent time (Perhaps 19 to 20 months.) on a Super Computer in order to test his latest trading theories. Prof. Richard Rundy of The University of Pennsylvaniaâs Wharton School, told The Daily Skirt that the worldâs computer experts fear that if aphexcoil gets within ten feet of any Super Computer, the entire planetâs network could crash.
For the first part the answer is yes: Without going into all the details the idea is this: If you have a single processor and perform a computation and then attempt to distribute the computation across n identical processors the limiting factors will be things like memory contention and I/O and communication bottlenecks (relative to the speed of the processors in working the problem). Usually the most important factor to the whole problem is whether the actual computation actually can be paralellized, that is broken up in a way that allows the distribution of problem is a uniform way across the processors. Some problems are inherently serial: that is they must be done sequentially. Others can be broken up into (hopefully) equal units and distributed. the speedups commonly run not as n times faster but something like n/Ln(n) which reflects the reduction from a linear speedup due to these issues. The speedup depends upon the suitability of the problem for distributed solution and your skill in building the algorithms ...
For the second part it sounds like you want to write a tradefeed / ticker processing system. This can be done without too much difficulty. Yes, here speed of analysis will be an issue but a lot can be done with a few multiprocessor intel boxes coupled with a fast I/o bus ......