Diagnosing performance problems in modern datacenters and distributed systems is challenging, as the root cause could be contained in any one of the system’s numerous components or, worse, could be a result of interactions among them. As distributed systems continue to increase in complexity, diagnosis tasks will only become more challenging. There is a need for a new class of diagnosis techniques capable of helping developers address problems in these distributed environments. As a step toward satisfying this need, this dissertation proposes a novel technique, called request-flow comparison, for automatically localizing the sources of performance changes from the myriad potential culprits in a distributed system to just a few potential ones. Request-flow comparison works by contrasting the workflow of how individual requests are serviced within and among every component of the distributed system between two periods: a non-problem period and a problem period. By identifying and ranking performance-affecting changes, request-flow comparison provides developers with promising starting points for their diagnosis efforts. Request workflows are obtained with less than 1% overhead via use of recently developed end-to-end tracing techniques. To demonstrate the utility of request-flow comparison in various distributed systems, this dissertation describes its implementation in a tool called Spectroscope and describes how Spectroscope was used to diagnose real, previously unsolved problems in the Ursa Minor distributed storage service and in select Google services. It also explores request-flow comparison’s applicability to the Hadoop File System. Via a 26-person user study, it identifies effective visualizations for presenting request-flow comparison’s results and further demonstrates that request-flow comparison helps developers quickly identify starting points for diagnosis.This dissertation also distills design choices that will maximize an end-to-end tracing infrastructure’s utility for diagnosis tasks and other use cases.

The intent of this study was to determine if men’s football and men’s basketball coaches at the university or college level utilize an assessment instrument when recruiting and evaluating potential student-athletes. Specifically studied through interviews were the characteristics that these coaches look for in successful and unsuccessful student-athletes, how they currently collect information during the recruitment period and the importance of collecting data on student-athletes. Swanson’s Performance Diagnosis Matrix and Human Capital Theory framed the research. The population for this study consisted of current Division IA men’s football and men’s basketball coaches in the Big 12 Conference. Prior to contacting the Big 12 coaches a pilot study was conducted at two Division IA Universities and with a former head football coach at a Big 12 Conference University. These interviews were instrumental in the final development of the questions used to interview the Big 12 Conference coaches. The participants were sent a letter asking for their participation in the study and then were contacted by phone to set up an interview. The interviews were conducted in the months of July, August, and September 2006 by phone. This study found that most coaches do not have or utilize an assessment instrument. Significant data showed coaches believe that the evaluation process of student-athletes is the most difficult and critical part of their job. Using emergent category designation I found seven themes (characteristics) of successful student-athletes, as indicated by the coaches: competitive, a hard worker, has a supportive family, is a leader, has good character, and is honest. I also found the themes (characteristics) of an unsuccessful student-athlete to be: undisciplined, lacks character, has an unstable family and is not competitive. The study helps to define through research and development an assessment instrument to more effectively define the needs of student-athletes prior to entering universities and coaches will have additional data for meeting the needs of student-athletes.

No description available.

This dissertation introduces, implements, and evaluates the novel concept of "Service Paths", which are system-level abstractions that capture and describe the dynamic dependencies between the different components of a distributed enterprise application. Service paths are dynamic because they capture the natural interactions between application services dynamically composed to offer some desired end user functionality. Service paths are distributed because such sets of services run on networked machines in distributed enterprise data centers. Service paths cross multiple levels of abstraction because they link end user application components like web browsers with system services like http providing communications with embedded services like hardware-supported data encryption. Service paths are system-level abstractions that are created without end user, application, or middleware input, but despite these facts, they are able to capture application-relevant performance metrics, including end-to-end latencies for client requests and the contributions to these latencies from application-level processes and from software/hardware resources like protocol stacks or network devices. Beyond conceiving of service paths and demonstrating their utility, this thesis makes three concrete technical contributions. First, we propose a set of signal analysis techniques called ``E2Eprof' that identify the service paths taken by different request classes across a distributed IT infrastructure and the time spent in each such path. It uses a novel algorithm called ``pathmap' that computes the correlation between the message arrival and departure timestamps at each participating node and detect dependencies among them. A second contribution is a system-level monitoring toolkit called ``SysProf', which captures monitoring information at different levels of granularity, ranging from tracking the system-level activities triggered by a single system call, to capturing the client-server interactions associated with a service paths, to characterizing the server resources consumed by sets of clients or client behaviors. The third contribution of the thesis is a publish-subscribe based monitoring data delivery framework called ``QMON'. QMON offers high levels of predictability for service delivery and supports utility-aware monitoring while also able to differentiate between different levels of service for monitoring, corresponding to the different classes of SLAs maintained for applications.