Figure 1. Multi-tier applications route data between clients and back-end storage. The Java component often acts as a hub that can be sampled to give insight into larger-scale performance problems. Figure 2. Data, sampled from a running JVMs, that we use for inferring the existence and nature of performance bottlenecks. Figure 4. Part of a 105-deep call stack, from the DayTrader bench- mark (a configuration of which is now part of the DaCapo Suite). Figure 3. From information that includes monitor states and stack sample context inferences, WAIT infers a Wait State for every sample. Table 2. Throughput perturbation versus sampling interval, from a document processing system running IBM’s 1.5 JVM. Table 1. The built-in mechanisms we use to sample the Java hub. Note that kill -3 does not terminate the signaled process, and 3 is the numeric code for SIGQUIT. Figure 5. Every concrete state, a sampled thread, maps to a pair of abstract states: a Wait State and a stack of Categories (one per frame). Every frame in a call stack has a default name based on the invoked package; e.g. com. MyBank.1login() would be named MyBank Code. Figure 6. The call stack is mapped to a stack of Categories, from which a primary Category (Database Communication) is chosen. Figure 8. The input concrete data model, after clustering and opti- mization. This example shows that cluster c1, which occurred 3592 times across all application samples, was invoked in thread t3 in the second application sample (sample index 2). Figure 7. The raw input from stack sampling provides this data for each sampled thread. Each monitor and thread has a hexadecimal address and a name (not shown). Table 6. Size of the data model communicated from server to the client’s browser. This data is a representative sample from actual IBM customer data that was submitted, by support personnel, to the WAIT service. Figure 12. WAIT Report: Lock Contention Figure 14. WAIT Report: Not Enough Load Figure 15. WAIT Report: Memory Leak Figure 16. WAIT Report: Database Bottleneck Figure 17. WAIT Report: Good Throughput but Filesystem Bottleneck
