Conclusion

The integration of TAU in the Cray XT3 environment was a successful endeavor and demonstrated the ability to incorporate system-specific features for performance observation in TAU's portable and configurable measurement facility. In particular, the leverage we gained from the memory introspection mechanisms in Cray's Catamount node kernel was significant and highlights how portable techniques provided by TAU can be improved by system-provided support. Certainly, we have witnessed such integration benefits on other platforms as well, but are especially satisfied with the range of integration specialization we achieved for the XT3. The work we have discussed is available as part of the latest TAU performance system distribution.