A Stall Metric to Track Communication Performance1

1 Certain commercial items may be identified but that does not imply recommendation or endorsement by NIST, nor does it imply that those items are necessarily the best available for the purpose. Abstract--- Probing the c ommunication p rotocol stack in Linux PC- based clusters to investigate err atic TCP/IP performance has led to a new metric, data stream stall, which is analogous to instruction stream stall in CPUs. Data stream stalling correlates well with unexpected throughput performance dips; the dips are usually due to delayed ACKs or questionable handling of them. We illustrate the use of this data stream stall metric by isolating and correcting the cause of these communication throughput dips in our version of Linux (2.0.29). The availability of this data stream stall metric would provide useful feedback to users by indicating deficient