Scaling up Mixed Workloads: a Battle of Data Freshness, Flexibility, and Scheduling

The common "one size does not fit all" paradigm isolates transactional and analytical workloads into separate, specialized database systems. Operational data is periodically replicated to a data warehouse for analytics. Competitiveness of enterprises today, however, depends on real-time reporting on operational data, necessitating an integration of transactional and analytical processing in a single database system. The mixed workload should be able to query and modify common data in a shared schema. The database needs to provide performance guarantees for transactional workloads, and, at the same time, efficiently evaluate complex analytical queries. In this paper, we share our analysis of the performance of two main-memory databases that support mixed workloads, SAP HANA and HyPer, while evaluating the mixed workload CH-benCHmark. By examining their similarities and differences, we identify the factors that affect performance while scaling the number of concurrent transactional and analytical clients. The three main factors are (a) data freshness, i.e., how recent is the data processed by analytical queries, (b) flexibility, i.e., restricting transactional features in order to increase optimization choices and enhance performance, and (c) scheduling, i.e., how the mixed workload utilizes resources. Specifically for scheduling, we show that the absence of workload management under cases of high concurrency leads to analytical workloads overwhelming the system and severely hurting the performance of transactional workloads.

Solid-State Storage and Work Sharing for Efficient Scaleup Data Analytics

Manoussos Gavriil Athanassoulis

Today, managing, storing and analyzing data continuously in order to gain additional insight is becoming commonplace. Data analytics engines have been traditionally optimized for read-only queries assuming that the main data reside on mechanical disks. The need for 24x7 operations in global markets and the rise of online and other quickly-reacting businesses make data freshness an additional design goal. Moreover, the increased requirements in information quality make semantic databases a key (often represented as graphs using the RDF data representation model). Last but not least, the performance requirements combined with the increasing amount of stored and managed data call for high-performance yet space-efficient access methods in order to support the desired concurrency and throughput. Innovative data management algorithms and careful use of the underlying hardware platform help us to address the aforementioned requirements. The volume of generated, stored and queried data is increasing exponentially, and new workloads often are comprised of time-generated data. At the same time the hardware is evolving with dramatic changes both in processing units and storage devices, where solid-state storage is becoming ubiquitous. In this thesis, we build workload-aware data access methods for data analytics - tailored for emerging time-generated workloads - which use solid-state storage, either (i) as an additional level in the memory hierarchy to enable real-time updates in a data analytics, or (ii) as standalone storage for applications involving support for knowledge-based data, and support for efficiently indexing archival and time-generated data. Building workload-aware and hardware-aware data management systems allows to increase their performance and to augment their functionality. The advancements in storage have led to a variety of storage devices with different characteristics (e.g., monetary cost, access times, durability, endurance, read performance vs. write performance), and the suitability of a method to an application depends on how it balances the different characteristics of the storage medium it uses. The data access methods proposed in this thesis - MaSM and BF-Tree - balance the benefits of solid-state storage and of traditional hard disks, and are suitable for time-generated data or datasets with similar organization, which include social, monitoring and archival applications. The study of work sharing in the context of data analytics paves the way to integrating shared database operators starting from shared scans to several data analytics engines, and the workload-aware physical data organization proposed for knowledge-based datasets - RDF-tuple - enables integration of diverse data sources into the same systems.

EPFL2014

Micro-architectural Analysis of Database Workloads

Utku Sirin

Database workloads have significantly evolved in the past twenty years. Traditional database systems that are mainly used to serve Online Transactional Processing (OLTP) workloads evolved into specialized database systems that are optimized for particular types of workloads. Data warehousing applications have led to Online Analytical Processing (OLAP) workloads and real-time analytical processing applications have led to Hybrid Transactional and Analytical Processing (HTAP) workloads. Similarly, modern hardware has significantly evolved in the past twenty years. Unicore, simple processors with megabytes of main memory have evolved into multi-core, power-limited processors with hundreds of gigabytes of main memory. Furthermore, the processors have complex micro-architectural features such as Single Instruction Multiple Data (SIMD) instructions and complex branch predictors. Advancements in processor technology have led to further evolution of database systems with novel system architectures and query processing paradigms. We present the micro-architectural behavior of modern database workloads on a modern processor for various categories of workloads and generations of database systems. We examine three main categories of database workloads and study them separately: OLTP, OLAP, and HTAP. We show that OLTP systems spend most of their execution time waiting for instruction-cache or data-cache misses, where the data-cache misses are due to the random data-accesses during the index lookup operation. While using an efficient index structure can significantly reduce the number of data-cache misses, the main micro-architectural bottleneck remains the data-cache misses due to the costly random data-accesses. Hence, OLTP systems should adopt techniques that mitigate the random data-accesses. OLAP systems spend most of their execution time in data-cache misses, where the data-cache misses are due to high pressure on the memory bandwidth for sequential-scan-heavy queries, and are due to random data-accesses for join-intensive queries. OLAP systems that follow tuple-at-a-time execution models efficiently use the CPU cycles. However, they require executing a significantly larger number of instructions hence are significantly slower than the systems that follow vector-at-a-time and compiled execution models. Therefore, OLAP systems should use efficient execution models and adopt techniques that mitigate data-cache misses. HTAP systems combine OLTP and OLAP systems into a single, unified system, where the OLTP and OLAP systems run on the same hardware and on the same data. Running on the same hardware results in hardware-level interference, where the OLTP throughput significantly drops due to the OLAP side sharing the hardware resources. Running on the same data results in increased OLAP query execution time due to the need for the OLAP side to process the fresh tuples generated by the OLTP side. Therefore, HTAP systems should adopt techniques that mitigate the hardware-level interference and should make sure that the OLAP side uses enough resources to minimize the increased query execution time.

EPFL2021

Micro-architectural Analysis of Database Workloads

Utku Sirin

EPFL2021

Solid-State Storage and Work Sharing for Efficient Scaleup Data Analytics

Manoussos Gavriil Athanassoulis

EPFL2014