Performance Characterization of Modern Databases on Out-of-Order CPUs

Memory Hierarchy Characterization of NoSQL Applications through Full-System Simulation

“…The characterization of these suites has been addressed by previous research works [27] [23] and their workloads are widely employed for microarchitecture research and evaluation. In many Big-data characterization works both PARSEC and SPEC benchmarks have usually been employed as a reference [9][13] [28] [11], highlighting the main differences between emerging scale-out applications and these suites. Following a similar approach, in this work we will also include the three aforementioned suites in the characterization process.…”

“…This database size has been chosen to fit into the 4GB Main Memory of the simulated machines (see section C). This is the recommended configuration in scale-out applications for optimal database performance [11]. Every NoSQL engine is configured making use of the default parameters described in YCSB repository.…”

“…Additionally, in order to limit simulation time the appropriate number of operations has also been validated. We have measured the MPKI evolution for a number of operations ranging from 10 to 10,000 (notice that the number of operations in [11] is 100,000). Three different Data Cache sizes (16Kb, 128Kb and 8Mb) were evaluated.…”

“…The authors in [11] make use of a similar software stack (YCSB as workload generator) and focus on the same specific kind of applications, that is, modern databases. By making use of profiling tools, performance and scalability of different commercial databases are analyzed.…”

“…For this reason, there is great interest in evaluating how this HW-SW distinctiveness can affect system behavior and performance. Numerous previous works [9][10] [11] [12] have made a remarkable effort in the characterization of Big-data applications. These works make use of benchmark suites covering broad application scenarios, in most cases running on current hardware.…”