Evaluating OSG Caching Infrastructure Frank Wrthwein OSG Executive
- Slides: 8
Evaluating OSG Caching Infrastructure Frank Würthwein OSG Executive Director Professor of Physics UCSD/SDSC
Asking the pertinent Questions • Does our caching infrastructure provide value to science? • Are there additional caching locations that we should consider deploying? 2
What Science? • Take the 6 science groups that are most heavily using the caching infrastructure • Do everything below for all of them, as much as applicable 3
What measurements? • Pick the top ten sites that these 6 customers ran on within the last 6 or 12 months. And do all evaluations only for those at first. • Define different types of measurements: Network related Job performance related Data volume related • Measure distribution of performance metrics for a given site for each science, and compare same science for different sites. • Compare same site for different sciences. 4
Network related measurements • RTT from WN to closest cache • Verify that closest cache choice is always the same … i. e. sanity check. • Can we measure bandwidth to WN from stash. CP logs ? • Can we measure dropped packets, or some other “quality of network” metric? 5
Job performance • CPU/wall time as measure of time wasted doing IO Make sure to also measure some CPU performance metric for the WNs to make sure that the efficiency ratio makes any sense !!! Probably need to normalize appropriately in some way. • Fraction of successful jobs out of total • What else? 6
Data Volume Related • For a given science, verify that the processing time and other metrics don’t radically depend on filesize that is being analyzed. • I can’t see why this matters any way other than as a crosscheck against systematic bias of other metrics as a function of filesize. 7
Final Comments • Consider this a first set of investigations • In all likelihood, we will find results that lead to more questions. • There are different source sof information that need to get correlated. • We may not have all the information we need to answer the questions we want to answer. 8
