How does the parallel version work?
The gene message database can often be quite large (as of the time of this writing, ours is 2.9GB). GrailEXP searches this database in partitions. In “serial mode”, it searches each partition one after the other (although the calls to BLAST can still be multithreaded). This only allows limited speedup, however, as a lot of the “aligning” is done by the GrailEXP search program itself. The true parallel version simply runs each GrailEXP search separately on different machines, where each task searches a single partition. This has several advantages over running BLAST multithreaded against a huge database, namely that: • The reading in of the database is done in parallel. • Each node uses substantially less memory. • The BLAST searches are less likely to hang when searching smaller databases. • (Most important) The GrailEXP alignment phase is run in parallel. Of course, you can also run BLAST multithreaded on partitions! You can do whatever you like; the program’s parallel search is fully
Related Questions
- When will Microsoft offer a parallel version of Microsoft SQL Server that runs on multiple servers at the same time for automatic load balancing and scalability?
- How do I run MrBayes MPI (parallel) version in the background on an SGI machine?
- What is the parallel MPI version of GARLI? Should I use it?
