Hardware counter driven on-the-fly request signatures

Abstract

Today's processors provide a rich source of statistical information on application execution through hardware counters. In this paper, we explore the utilization of these statistics as request signatures in server applications for identifying requests and inferring highlevel request properties (e.g., CPU and I/O resource needs). Our key finding is that effective request signatures may be constructed using a small amount of hardware statistics while the request is still in an early stage of its execution. Such on-the-fly request identification and property inference allow guided operating system adaptation at request granularity (e.g., resource-aware request scheduling and on-the-fly request classification). We address the challenges of selecting hardware counter metrics for signature construction and providing necessary operating system support for per-request statistics management. Our implementation in the Linux 2.6.10 kernel suggests that our approach requires low overhead suitable for runtime deployment. Our on-the-fly request resource consumption inference (averaging 7%, 3%, 20%, and 41% prediction errors for four server workloads, TPC-C, TPC-H, J2EE-based RUBiS, and a trace-driven index search, respectively) is much more accurate than the online running-average based prediction (73-82% errors). Its use for resource-aware request scheduling results in a 15-70% response time reduction for three CPU-bound applications. Its use for on-the-fly request classification and anomaly detection exhibits high accuracy for the TPC-H workload with synthetically generated anomalous requests following a typical SQL-injection attack pattern.

Pruning hardware evaluation space via correlation-driven application similarity analysis

“…The availability of dedicated performance monitoring unit on modern micro-architectures [20,21] enables the collection of run-time application characteristics [22], with negligible overhead. The Sampling Engine collects the data for CPI and resource stalls as per the schema shown in Figure 2.…”

Empirical examination of a collaborative web application

“…We expect that the importance of collaborative web applications will rapidly rise given their surging popularity and vitality through direct user creation. The goal of this work is to raise awareness on the need for collaborative web application benchmarks, particularly for the purpose of evaluating computer system support for online services [3,6,15,[18][19][20].…”

“…All deployed application/benchmarks are CPU-bound. The server operating system is Linux 2.6.10 with an augmented request context maintenance and event tracing framework [15]. This framework allows us to collect perrequest event traces for multi-component server applications.…”

“…Although some user requests may result in persistent state changes on the server side, such changes often exert very small effects on system-level application execution behaviors. Due to the large dependence on central content generation sources in these applications, the resource requirements and other characteristics for processing user requests are typically well clustered into small numbers of groups [3,6,15]. Widely used online benchmarks such as TPC benchmarks (e.g., TPC-C, TPC-H, TPC-W) [22], SPEC online benchmarks (e.g., SPECweb, SPECjbb, SPECjAppServer) [16], and J2EE-based multi-component benchmarks (e.g., RUBiS [13], StockOnline [21]) all reflect such request behavior patterns.…”

Power and energy containers for multicore servers

“…Building on our past work [3,5] on per-request behavior characterization in a multicore server, we address the additional challenges of per-request power/energy accounting and management. Request executions in a concurrent, multi-stage server contain finegrained activities with frequent context switches.…”