Alexander Zeier

SIMD-scan: ultra fast in-memory table scan using on-chip vector processing units

The availability of huge system memory, even on standard servers, generated a lot of interest in main memory database engines. In data warehouse systems, highly compressed column-oriented data structures are quite prominent. In order to scale with the data volume and the system load, many of these systems are highly distributed with a shared-nothing approach. The fundamental principle of all systems is a full table scan over one or multiple compressed columns. Recent research proposed different techniques to speedup table scans like intelligent compression or using an additional hardware such as graphic cards or FPGAs. In this paper, we show that utilizing the embedded Vector Processing Units (VPUs) found in standard superscalar processors can speed up the performance of mainmemory full table scan by factors. This is achieved without changing the hardware architecture and thereby without additional power consumption. Moreover, as on-chip VPUs directly access the systems RAM, no additional costly copy operations are needed for using the new SIMD-scan approach in standard main memory database engines. Therefore, we propose this scan approach to be used as the standard scan operator for compressed column-oriented main memory storage. We then discuss how well our solution scales with the number of processor cores; consequently, to what degree it can be applied in multi-threaded environments. To verify the feasibility of our approach, we implemented the proposed techniques on a modern Intel multi-core processor using Intel® Streaming SIMD Extensions (Intel® SSE). In addition, we integrated the new SIMD-scan approach into SAP® Netweaver® Business Warehouse Accelerator. We conclude with describing the performance benefits of using our approach for processing and scanning compressed data using VPUs in column-oriented main memory database systems.

A Hybrid Row-Column OLTP Database Architecture for Operational Reporting

Operational reporting differs from informational reporting in that its scope is on day-to-day operations and thus requires data on the detail of individual transactions. It is often not desirable to maintain data on such detailed level in the data warehouse, due to both exploding size of the warehouse and the update frequency required for operational reports. Using an ODS as the source for operational reporting exhibits a similar information latency.

Towards Analytics-as-a-Service Using an In-Memory Column Database

For traditional data warehouses, mostly large and expensive server and storage systems are used. For small- and medium size companies, it is often too expensive to implement and run such systems. Given this situation, the SaaS model comes in handy, since these companies might opt to run their OLAP as a service. The challenge is then for the analytics service provider to minimize TCO by consolidating as many tenants onto as few servers as possible, a technique often referred to as multi-tenancy.

A case for online mixed workload processing

Database systems in the context of business data processing are segmented into two categories: those intended for online transaction processing (OLTP) and those for online analytical processing (OLAP). Over the last 15 years, database management system (DBMS) proposals directly addressing one of those categories were most represented in terms of academic publications and variety of commercial products in the domain of enterprise computing.n In contrast, the most innovative DBMS proposals in this century were invented not by addressing a well-known category but by following a methodology that purely focuses on the application characteristics as practiced by Amazon or Google. This paper applies a part of that methodology to the field of enterprise applications in order to evaluate to what extend they are covered by the categories OLTP and OLAP. The evaluation shows that there are enterprise applications that reveal a mix of those characteristics which are usually exclusively associated either with OLTP or with OLAP and therefore cannot be addressed adequately by traditional DBMS.n The paper contributes by pointing out that those applications cause an online mixed workload and by explaining what properties a corresponding specialized DBMS should have and how this category of enterprise applications could benefit from it.

An Instrument for Real-Time Design Interaction Capture and Analysis

How do designers leverage information and communication technology to collaborate with team partners and other process participants? Given the increasingly complex, distributed, and virtual setups of design environments and processes, answering this question is challenging. At HPI, we have developed computational data collection and analysis techniques to improve the efficiency and range of observations in technology-enabled design spaces. Using our software, we were able to capture and evaluate complex characteristics of online interactions in distributed design teams at quasi real-time. Besides new insights into the communication behavior of design teams, it could be demonstrated that communication activity signatures of high-performance design teams are significantly different than those of low-performance teams. The combination of new techniques along with quantifiable performance metrics provides a stable foundation for real-time design team diagnostics.

Cache-conscious data placement in an in-memory key-value store

Key-value stores which keep the data entirely in main memory can serve applications whose performance criteria cannot be met by disk-based key-value stores. This paper evaluates the performance implications of cache-conscious data placement in an in-memory key-value store by examining how many values have to be stored consecutively in blocks in order to fully exploit memory locality during bandwidth-bound operations. We contribute by introducing a random block traversal main memory access pattern, by describing the corresponding memory access costs as well as by formally and experimentally deriving the correlation between block size and throughput. Our calculations and experiments vary the value and block sizes as well as their placement in the memory and derive their impact on cache-misses throughout the different memory hierarchies, the ability to prefetch data, and the number of needed CPU cycles to perform a certain set of data operations. The paper closes with the insight that a block-wise grouping of relatively few key-value pairs increases the throughput up to a factor six and with a discussion which implications a block-wise grouping of data has on the system design key-value store.

Towards a Shared Platform for Virtual Collaboration Monitoring in Design Research

Prior applications of a system to monitor IT-mediated communication activities of design teams provided new insights into the collaboration behavior during the early phases of concept creation and prototyping. We now take our approach to the next level by sketching an architecture for a platform that aims to establish ‘out- of-the-box’ monitoring capabilities for virtual team environments and to facilitate the sharing and evaluation of recorded activities within a larger research community. To further demonstrate the flexibility and applicability of our instrument, we present results and experiences gained from a recently conducted observation of software engineering teams. Our vision is a common service for capturing and analyzing virtual collaboration activities that promotes comparative research and team diagnostics in engineering design.

Leveraging multi-core CPUs in the context of demand planning

Over the last couple of years a paradigm shift in CPU development has been happening: away from increasing the clock rate towards increasing the numbers of cores per CPU. In order to leverage the resulting new architectures, nowadays business applications have to be rewritten or at least adapted. This paper describes the most characteristic operations in the context of demand planning which is a crucial part of supply chain management. Furthermore, it presents different strategies how those operations can be distributed onto several cores. The different strategies were prototypically implemented and measured on different systems (including Intel Core and Nehalem Microarchitecture) in order to determine their efficiency. The paper concludes with a comparison of the different strategies, identifies the most efficient one, and describes the potential performance gain in demand planning by leveraging multi-core CPUs.