No abstract

Online Internet applications see dynamic workloads that fluctuate over multiple time scales. This paper argues that the non-stationarity in Internet application workloads, which causes the request mix to change over time, can have a significant impact on the overall processing demands imposed on data center servers. We propose a novel mix-aware dynamic provisioning technique that handles both the non-stationarity in the workload as well as changes in request volumes when allocating server capacity in Internet data centers. Our technique employs the k-means clustering algorithm to automatically determine the workload mix and a queuing model to predict the server capacity for a given workload mix. We implement a prototype provisioning system that incorporates our technique and experimentally evaluate its efficacy on a laboratory Linux data center running the TPC-W web benchmark. Our results show that our k-means clustering technique accurately captures workload mix changes in Internet applications. We also demonstrate that mix-aware dynamic provisioning eliminates SLA violations due to under-provisioning with non-stationary web workloads, and that it offers a better resource usage by reducing over-provisioning when compared to a baseline provisioning approach that only reacts to workload volume changes. We also present a case study of our provisioning approach on Amazon's EC2 cloud platform.

Continuous "always-on" monitoring is beneficial for a number of applications, but potentially imposes a high load in terms of communication, storage and power consumption when a large number of variables need to be monitored. We introduce two new filtering techniques, swing filters and slide filters, that represent within a prescribed precision a time-varying numerical signal by a piecewise linear function, consisting of connected line segments for swing filters and (mostly) disconnected line segments for slide filters. We demonstrate the effectiveness of swing and slide filters in terms of their compression power by applying them to a reallife data set plus a variety of synthetic data sets. For nearly all combinations of signal behavior and precision requirements, the proposed techniques outperform the earlier approaches for online filtering in terms of data reduction. The slide filter, in particular, consistently dominates all other filters, with up to twofold improvement over the best of the previous techniques.

On-line services are making increasing use of dynamically generated Web content. Serving dynamic content is more complex than serving static content. Besides a Web server, it typically involves a server-side application and a database to generate and store the dynamic content. A number of standard mechanisms have evolved to generate dynamic content. We evaluate three specific mechanisms in common use: PHP, Java servlets, and Enterprise Java Beans (EJB). These mechanisms represent three different architectures for generating dynamic content. PHP scripts are tied to the Web server and require writing explicit database queries. Java servlets execute in a different process from the Web server, allowing them to be located on a separate machine for better load balancing. The database queries are written explicitly, as in PHP, but in certain circumstances the Java synchronization primitives can be used to perform locking, reducing database lock contention and the amount of communication between servlets and the database. Enterprise Java Beans (EJB) provide several services and facilities. In particular, many of the database queries can be generated automatically. We measure the performance of these three architectures using two application benchmarks: an online bookstore and an auction site. These benchmarks represent common applications for dynamic content and stress different parts of a dynamic content Web server. The auction site stresses the server front-end, while the online bookstore stresses the server back-end. For all measurements, we use widely available open-source software (the Apache Web server, Tomcat servlet engine, JOnAS EJB server, and MySQL relational database). While Java servlets are less efficient than PHP, their ability to execute on a different machine from the Web server and their ability to perform synchronization leads to better performance when the front-end is the bottleneck or when there is database lock contention. EJB facilities and services come at the cost of lower performance than both PHP and Java servlets.