Monday, March 24, 2008
taskq
We found a new openmp Pragma ''taskq" which works well with recursive functions. The recursive functions are naturally parallelized with the workqueuing model. We tried a few examples and it works fine. We are modifying our programs with taskq.
Tuesday, March 11, 2008
Memory Leak and Niagara
We used Valgrind to check the original code. One memory leak is found so far. We have fixed it. We will keep checking other possible problems.
As to Niagara, we run the original code without multi-threading and the rulecahce update time of one trace file is cut to half. It looks Niagara already did some optimization. We will try other trace files. The results from multi-threaded code is not very good. We will keep looking for fixes and other potential optimizations.
As to Niagara, we run the original code without multi-threading and the rulecahce update time of one trace file is cut to half. It looks Niagara already did some optimization. We will try other trace files. The results from multi-threaded code is not very good. We will keep looking for fixes and other potential optimizations.
Tuesday, March 4, 2008
Consistency and Memory Errors
We tested the code with the changes we made and the original with a small ruleset and the output from the decision for the packets has been exactly the same.
We are trying now to verify this using the whole ruleset but we are running into problems. On the new machine with a GCC version that supports openmp our program (the original version) eats up all the memory until it crashes. We ran the original code also on the baleen machine and it runs fine. We suspect the different versions of GCC are to blame and we are working with Mike to come up with a solution for this.
We are trying now to verify this using the whole ruleset but we are running into problems. On the new machine with a GCC version that supports openmp our program (the original version) eats up all the memory until it crashes. We ran the original code also on the baleen machine and it runs fine. We suspect the different versions of GCC are to blame and we are working with Mike to come up with a solution for this.
Subscribe to:
Posts (Atom)