Scalability in HPC Computing Environments
TotalView provides features and performance enhancements for scalable debugging in today’s HPC computing environments, and no special configuration or action is necessary on your part to take advantage of TotalView’s scalability abilities.
This chapter details TotalView’s features and configurations related to scalability, as follows:
The Processes and Threads View. This view aggregates program state so that it can display quickly and is easy to understand. Scalability Configuration Settings. Depending on your needs, you might want to set specific configuration variables that enable scalable debugging operations. MRNet Configuration Settings. TotalView uses MRNet, a tree-based overlay network, for scalable communication. TotalView is preconfigured for scalability, but in some situations you may want to change MRNet's configuration. dstatus and dwhere command options. These options provide aggregated views of various process and thread properties. (See the -group_by option in the Reference Guide entries for these commands.)Compressed process/thread list. The ptlist compactly displays the set of processes and threads that have been aggregated together.Compressed List Syntax (ptlist) | “Compressed List Syntax (ptlist)” in the dstatus entry of the Reference Guide |
dstatus -group_by and dwhere-group_by options | dstatus and dwhere in the Reference Guide |
