Cache Pirating: Measuring the Curse of the Shared Cache

Abstract

We present a low-overhead method for accurately measuring application performance (CPI) and off-chip bandwidth (GB/s) as a function of its the available shared cache capacity, on real hardware, with no modifications to the application or operating system. We accomplish this by co-running a Pirate application that "steals" cache space with the Target application. By adjusting how much space the Pirate steals during the Target's execution, and using hardware performance counters to record the Target's performance, we can accurately and efficiently capture performance data for the Target application as a function of its available shared cache. At the same time we use performance counters to monitor the Pirate to ensure that it is successfully stealing the desired amount of cache.To evaluate this approach, we show that 1) the cache available to the Target behaves as expected, 2) the Pirate steals the desired amount of cache, and 3) the Pirate does not impact the Target's performance. As a result, we are able to accurately measure the Target's performance while stealing between 0MB and an average of 6.1MB of the 8MB of cache on our Nehalem based test system with an average measurement overhead of only 5.5%.

Efficient techniques for predicting cache sharing and throughput

“…This input data consist of the applications' fetch and hit rates, IPCs, and hit ratios as a function of their cache allocation, and can be acquired with low overhead on modern multicore machines [6]. This low-overhead data is in contrast to many existing methods for modeling cache sharing which rely on expensive data such as stack distance traces [2,4,3,17].…”

Modeling performance variation due to cache sharing

“…Our method combines and extends three existing pieces of infrastructure: a cache sharing model [16], a lowoverhead cache analysis tool [7], and a phase detection framework [19]. In this section, we describe the different pieces and how we extend them.…”

“…Traditionally, such profiles have been generated through simulation, but such an approach is slow and it is difficult to build accurate simulators for modern processor pipelines and memory systems. Instead, we use Cache Pirating [7] to collect the data. Cache Pirating solves both problems by measuring how an application behaves as a function of cache size on the target machine with very low overhead.…”

“…These profiles contain information about how the miss rate (misses per cycle) and hit rate (hits per cycle) vary for an application as a function of cache size. We use the Cache Pirating [7] technique (discussed below) to capture the model's input data.…”

“…This allows us to efficiently predict the performance and bandwidth requirements of mixed workloads. In addition, the input data to the cache model is captured using low-overhead profiling [7] of each application running in isolation. This means that only a small number of profiling runs need to be done on the target machine.…”