Cipres Web Portal

---

MrBayes 3.1.2 on TG

MrBayes on TG provides a familiar MrBayes interface, but submits to Trestles, a large NSF TeraGrid Resource. The Trestles resource will run your MrBayes job, and return the results just like the original MrBayes interface.

Advantages: MrBayes on TG can handle larger job runs than our original MrBayes interface. The Trestles machine allows your job to run for up to 7 days. The interface will run the job using parallel MrBayes, so if nruns=2 or more and nchains = 4 or more (default conditions) run speed is 3 - 4.5 times faster than the CIPRES cluster. In other words, the computation may be equivalent to 21 -28 days on the (sequential) CIPRES Cluster.

Disadvantages:The queue times on Teragrid resources vary according to traffic and according to the size of your job (shorter jobs typically have shorter queue waits). Sometimes TeraGrid resources go down for preventative maintenance or due to unforeseen circumstances. At present, this will cause your job to fail, and you will have to restart it. There is no restart capability at present.

Click this icon to check the traffic on TeraGrid:
Click this icon to check the Teragrid maintenance schedule:

How to use it: The interface for MrBayes on TG differs from the original interface by presenting a form that requires the user to enter the number of runs and chains, and a maximum wall clock time for each job. This will determine the job priority in the queue, help identify the number of processors required, and set a time for job termination (whether the run is finished or not). If you set a value of 0.5 hr or less, your job will run in a testing queue, which generally has a much shorter wait time than the production queue.

Should you use it? If you have jobs that run more than two days on the CIPRES cluster, you should definitely try to use MrBayes on Trestles.

Known Issues: MrBayes 3.1.2 has a known issue where the job can "runaway", creating a large log file, and making no progress. At present we beleive this is triggered by runs that have low generation counts (< 5,000 gens). We have tried to compensate for this by preventing users from submitting jobs less than 5,000 generations. Should you experience, or think you experienced a runaway job, please contact us. We expect this issue to be resolved in the next release of MrBayes (MB V 3.2.0)..

If there is a tool or a feature you need, let us know.