CPU and cache efficient management of memory-resident databases

Abstract

Abstract-Memory-Resident Database Management Systems (MRDBMS) have to be optimized for two resources: CPU cycles and memory bandwidth. To optimize for bandwidth in mixed OLTP/OLAP scenarios, the hybrid or Partially Decomposed Storage Model (PDSM) has been proposed. However, in current implementations, bandwidth savings achieved by partial decomposition come at increased CPU costs. To achieve the aspired bandwidth savings without sacrificing CPU efficiency, we combine partially decomposed storage with Just-in-Time (JiT) compilation of queries, thus eliminating CPU inefficient function calls. Since existing cost based optimization components are not designed for JiT-compiled query execution, we also develop a novel approach to cost modeling and subsequent storage layout optimization. Our evaluation shows that the JiT-based processor maintains the bandwidth savings of previously presented hybrid query processors but outperforms them by two orders of magnitude due to increased CPU efficiency.

In-cache query co-processing on coupled CPU-GPU architectures

“…Particularly, all previous studies assign homogeneous workloads to the CPU and the GPU. From these observations, the GPU is severely degraded by the memory stalls, which are usually a major performance factor for databases [27,19]. Homogeneous workload distribution still causes excessive memory stalls on the GPU, despite the fine-grained and collaborative improvements in the previous studies [19,38,21,7].…”

Non-invasive progressive optimization for in-memory databases

“…Furthermore, the generic cost model introduced by Manegold et al [13] allows us to model the cache accesses for other relational operators such as joins or sorts by combining atomic access patterns. Figure 2 shows L3 accesses for an increasing selectivity described by Pirk et al [17]. The main reason for this behavior is the high number of random misses for small selectivities.…”

“…Each subsequent predicate introduces a sequential scan with conditional read pattern which induces cache accesses depending on the selectivity of the previous predicate. We refer to Pirk et al [17] for a detailed description of this model. Furthermore, the generic cost model introduced by Manegold et al [13] allows us to model the cache accesses for other relational operators such as joins or sorts by combining atomic access patterns.…”

“…The extension of the generic cost model by Pirk et al [17] allows us to model the cache accesses of different PEOs. We estimate induced cache accesses of a multi-selection query by exploiting two patterns.…”

“…Based on an extended evaluation of the cost model on modern CPUs, we modify the cost model by Pirk et al [17] to double count the number of random misses which leads to more precise estimations. This modification models the effect, that a random cache miss induces one cache access for the cache line that was predicted but not used and one cache line access for the actually used cache line.…”