Main Memory Database Research Articles

All-IP 이동 통신망에서 HSS 시스템의 효과적인 백업과 실시간 이중화 기법

디지털 이동통신 서비스의 가입자 정보를 실시간으로 관리하는 HSS(Home Subscriber Server) 시스템은 사용자의 콜처리를 위한 기본 정보뿐만 아니라 사용자 인증과 운영 등에 관한 추가적인 정보를 제어하기 위해 가입자 데이터베이스를 주기억장치에 상주시켜 운영하는 주기억 장치 데이터베이스(main memory database system)를 필요로 한다. 그러나, 신뢰성과 안정성이 중요시되는 HSS-DBS의 경우, 초기 로딩시간이 길고 메모리 손실시 그 영향이 훨씬 크기 때문에 보다 안전한 데이터의 저장 방식과 백업 기법이 필요하다. 본 논문에서는 HSS 시스템의 안전성과 성능을 고려한 효율적인 백업 이중화 기법을 제시하였다. 제안한 쉐도잉 백업 기법을 통해 위치 등록으로 인한 실시간 백업의 과부하를 최소화하고 일정 시간 간격을 두고 지연 회복시킴으로써 오류 발생 시 신속하게 복구할 수 있으며 원격 서버에 실시간으로 백업하는 백업 이중화 기법을 통해 안정적인 시스템 운영을 보장한다. An HSS(Home Subscriber Server) system requires a main-memory database on main-memory unit for the real-tine management of the subscriber information in the mobile communication service, in that the system controls not only basic data for handling calls of users, but also additional service data related to user authentication and operational data. Nonetheless, HSS-DBS system, requiring the reliability and stability, need more secure data store method and a back-up technique because the system have a long startup time and the big problem on the failures of main-memory. This paper proposes an efficient back-up replication technique, on the basis of enhancing the stability and performance of HSS system. The proposed shadowing back-up technique adopting the delayed recovery process, can help minimize the real-time back-up overloads by location registration, while the proposed backup replication method enables more stable system operations with replicating the data to remote server in real time.

Main Memory Database TPC-H Workload Characterization on Modern Processor

Ailamaki等人1999年研究了数据库管理系统(database management system,简称DBMS)在处理器上的时间开销分解.此后,相关研究集中在分析DBMS在处理器上的瓶颈.但这些研究工作均是在磁盘数据库DRDBs(disk resident databases)上开展的,而且都是分析DBMS上的TPC-C类负载.然而,随着硬件技术的进步,现代计算机的多级缓存结构(memory hierarchy)在逐渐地"上移".例如,容量越来越大的芯片内缓存(on-chip caches)和芯片外缓存(off-chip caches),容量越来越大的RAM,Flash Memory等等.为此,处理器负载分析的研究工作也应随之"上移".研究内存数据MMDBs(main memory resident databases)在计算密集型负载下的处理器行为特性.由于磁盘数据库的主要性能瓶颈是磁盘I/O,因而可以用索引、压缩等技术进行优化;然而,内存数据库的性能瓶颈却在于处理器和内存之间的数据交换.针对这一问题,首先分析了磁盘数据库和内存数据库在TPC-H负载下处理器性能瓶颈的差异,并给出了一些优化建议,提出了通过预取的优化方法.其次,通过实验比较了不同存储体系结构(行存储与列存储)对处理器利用率的差异,并探索了下一代内存数据库体系结构方面的解决方案.此外,还研究了索引结构对处理器多级缓存的影响,并给出了索引的优化建议.最后,提出一个微测试集用于评估内存数据库在DSS(decision support system)负载下处理器的性能及行为特性.研究结果会对运行于下一代处理器上的内存数据库体系结构设计和性能优化提供一定的实验依据.;In 1999, the research of database systems' execution time breakdown on modern computer platforms has been analyzed by Ailamaki, et al. The primary motivation of these studies is to improve the performance of Disk Resident Databases (DRDBs), which form the main stream of database systems until now. The typical benchmark used in those studies is TPC-C. However, continuing hardware advancements have moved-up on the memory hierarchy, such as the larger and larger on-chip and off-chip caches, the steadily increasing RAM space, and the commercial availability of huge flash memory (solid-state disk) on top of regular disk, etc. To reflect such a trend, the target of workload characterization research along the memory hierarchy is also studied. This paper focuses on Main Memory Databases (MMDBs), and the TPC-H benchmark. Unlike the performance of DRDB which is I/O bound and may be optimized by high-level mechanisms such as indexing, the performance of MMDB is basically CPU and memory bound. In this study, the paper first compares the execution time breakdown of DRDB and MMDB, and the paper proposes an optimize strategy to optimize the memory resident aggregate. Then, the paper explores the difference between column-oriented and row-oriented storage models in CPU and cache utilization. Furthermore, the paper measures performance of MMDBs on different generational CPUs. In addition, the paper analyzes the index influence and gives a strategy for main memory database index optimization. Finally, the paper analyzes each query in the full TPC-H benchmark in detail, and obtains systematic results, which help design micro-benchmarks for further analysis of CPU cache stall. Results of this study are expected to benefit the performance optimization of MMDBs, and the architecture design memory-oriented databases of the next generation.

CC-GiST: 임의의 캐시 인식 검색 트리를 효율적으로 구현하기 위한 일반화된 프레임워크

최근 메인 메모리 가격이 하락하고 용량이 크게 증가함에 따라 메인 메모리 데이터베이스에 기반한 응용이 급격히 증가하고 있다. 캐시 미스 (cache miss)는 CPU에서 액세스하고자 하는 데이터가 캐시에 존재하지 않아 메모리로부터 읽어 들이는 과정이며, 메인 메모리 데이터베이스의 성능 감소의 중요한 원인이다. 메인 메모리 데이터베이스에서의 캐시 미스를 줄이고 캐시를 최대한 활용하기 위하여 여러 가지 캐시 인식 트리들(cache conscious trees)이 제안되었다. 이러한 캐시 인식 트리들은 각각 특성이 다르므로 하나의 응용에서 둘 이상의 캐시 인식 트리들이 동시에 관리될 수 있다. 또한, 만약 기존의 캐시 인식 트리가 응용에서의 요구를 만족시키지 못하면 새로운 캐시 인식 트리를 구현하여야 한다. 본 논문에서는 캐시 인식하는 일반화된 검색 트리(Cache-Conscious Generalized Search Tree, CC-GiST)를 제안한다. CC-GiST는 디스크 기반의 일반화된 검색 트리 (Generalized Search Tree, GiST) [HNP95]를 캐시 인식하도록 확장한 것이며, 포인터 압축(pointer compression)과 키 압축(key compression) 기법을 비롯하여 임의의 캐시 인식 트리의 공통적인 기능 및 알고리즘들을 동시에 제공한다. CC-GiST를 기반으로 특정 캐시 인식트리를 구현하려면 그 트리에 해당된 기능만을 구현하면 된다. 본 논문에서는 CC-GiST를 기반으로 기존의 대표적인 캐시 인식 트리인 CSB+-트리, pkB-트리, CR-트리를 구현하는 방법을 기술한다. CC-GiST를 이용함에 따라 메인 메모리 데이터베이스 응용에서 여러 개의 캐시 인식 트리를 관리하는 번거로움에서 벗어날 수 있고, 응용의 요구에 따른 새로운 캐시 인식 트리를 최소한의 노력으로 효율적으로 구현할 수 있다. According to recent rapid price drop and capacity growth of main memory, the number of applications on main memory databases is dramatically increasing. Cache miss, which means a phenomenon that the data required by CPU is not resident in cache and is accessed from main memory, is one of the major causes of performance degradation of main memory databases. Several cache-conscious trees have been proposed for reducing cache miss and making the most use of cache in main memory databases. Since each cache-conscious tree has its own unique features, more than one cache-conscious tree can be used in a single application depending on the application's requirement. Moreover, if there is no existing cache-conscious tree that satisfies the application's requirement, we should implement a new cache-conscious tree only for the application's sake. In this paper, we propose the cache-conscious generalized search tree (CC-GiST). The CC-GiST is an extension of the disk-based generalized search tree (GiST) [HNP95] to be tache-conscious, and provides the entire common features and algorithms in the existing cache-conscious trees including pointer compression and key compression techniques. For implementing a cache-conscious tree based on the CC-GiST proposed in this paper, one should implement only a few functions specific to the cache-conscious tree. We show how to implement the most representative cache-conscious trees such as the CSB+-tree, the pkB-tree, and the CR-tree based on the CC-GiST. The CC-GiST eliminates the troublesomeness caused by managing mire than one cache-conscious tree in an application, and provides a framework for efficiently implementing arbitrary cache-conscious trees with new features.

A Real-Time Dynamic Database Crash Recovery Strategy

Distributed real-time main memory databases are usually applied in some time-critical applications. For these applications,rapid and efficient recovery in the event of site crash is very important. Firstly,through analyzing the recovery requirements of failure in distributed real-time main memory databases,the recovery correctness criteria of distributed real-time main memory database are presented. Then,a real-time dynamic crash recovery scheme based on log (RTDCRS) is presented,and the correctness of RTDCRS is proved. RTDCRS adopts the real-time logging scheme integrating the characteristics of partitioned logging,ephemeral logging and uses nonvolatile high-speed store as logging storage area in order to reduce the logging cost during the normal running. After a site crash,a dynamic recovery strategy based on classification recovery idea is adopted to decrease the downtime. Performance tests and evaluations results show that RTDCRS has better performances than the traditional recovery schemes in two aspects: The missing deadlines ratio of transactions and the time of system denying services after crashes.

A dynamic crash recovery scheme for distributed real-time database systems

Recovery performance in the event of failures is very important for distributed real-time database systems. This paper presents a time-cognizant logging-based crash recovery scheme (TCLCRS) that aims at distributed real-time databases, which adopts a main memory database as its ground support, hi our scheme, each site maintains a real-time log for local transactions and the sub-transactions, which execute at the site, and execute local checkpointing independently. Log records are stored in non-volatile high-speed store, which is divided into four different partitions based on transaction classes. During restart recovery after a site crash, partitioned crash recovery strategy is adopted to ensure that the site can be brought up before the entire local secondary database is reloaded in main memory. The partitioned crash recovery strategy not only guarantees the internal consistency to be recovered, but also guarantee the temporal consistency and recovery of the sates of physical world influenced by uncommitted transactions. Combined with two-phase commit protocol, TCLCRS can guarantee failure atomicity of distributed real-time transactions.

A Partition Fuzzy Checkpointing Strategy for Real-Time Main Memory Databases

A Partition Fuzzy Checkpointing Strategy for Real-Time Main Memory Databases

위치기반 서비스를 위한 다중레벨 DBMS에 질의 분류 컴포넌트의 설계 및 구현

현재 위치기반 서비스를 제공하기 위하여 다양한 시스템들이 사용되고 있다. 그러나 기존의 시스템들은 상당히 많은 사용자들에게 빠른 서비스를 제공하기에는 적합하지가 않다. 이러한 문제점을 해결하기 위하여 빠른 데이터 처리와 대용량의 데이터 관리를 동시에 지원하는 다중레벨 DBMS를 사용하여야 한다. 스냅샷을 갖는 다중레벨 DBMS는 디스크에 모든 데이터를 가지고 있으며, 빠른 처리를 요구하는 데이터는 스냅샷의 형태로 메인메모리 데이터베이스에서 관리한다. 이 시스템의 성능을 최적화하여 위치기반 서비스를 제공하기 위해서는 스냅샷에 존재하는 데이터를 효율적으로 사용할 수 있도록 질의를 분류하는 컴포넌트가 필요하다. 본 논문에서는 위치기반 서비스를 위한 다중레벨 DBMS에서 질의 분류 컴포넌트를 설계하고 구현한다. 제안된 컴포넌트는 입력된 질의를 메모리 질의, 디스크 질의, 하이브리드 질의로 분류하여 스냅샷 사용율을 높이고, 스냅샷의 일부분을 사용할 수 있도록 질의의 비공간과 공간 필터 조건을 분할하는 메커니즘을 사용하였다. 따라서, 제안된 컴포넌트는 효율적인 질의 분류를 통하여 스냅샷을 최대한 이용함으로써 시스템의 성능을 향상시킨다. Various systems are used to provide the location based services. But, the existing systems have some problems which have difficulties in dealing with faster services for above million people. In order to solve it, a multi-level DBMS which supports both fast data processing and large data management support should be used. The multi-level DBMS with snapshots has all the data existing in disk database and the data which are required to be processed for fast processing are managed in main memory database as snapshots. To optimize performance of this system for location based services, the query classification component which classifies the queries for efficient snapshot usage is needed. In this paper, the query classification component in multi-level DBMS for location based services is designed and implemented. The proposed component classifies queries into three types: (1) memory query, (2) disk query, (3) hybrid query, and increases the rate of snapshot usage. In addition, it applies division mechanisms which divide aspatial and spatial filter condition for partial snapshot usage. Hence, the proposed component enhances system performance by maximizing the usage of snapshot as a result of the efficient query classification.

Tmr-트리 : 주기억 데이터베이스에서 효율적인 공간 색인 기법

최근 들어 계속되는 램 가격 하락으로 인해 대용량의 램을 사용하는 주기억 데이터베이스 시스템의 구축이 실현 가능하게 되었다. 그러나 기존의 디스크 기반 공간 색인 기법은 디스크 접근 시간만을 주로 고려하기 때문에, 주기억 색인 기법으로 디스크 기반 색인 기법을 직접적으로 적용시키는 것은 부적절하다. 주기억 장치 색인 기법은 모든 색인 노드들이 주기억 장치에 상주하기 때문에 노드에 대한 접근 시간이 디스크 기반 기법에 비해 상당히 미미하고, 결국 효율적인 색인 기법을 위해서는 노드 접근시간 뿐만 아니라 노드내의 키 비교시간을 고려해야 한다. 이러한 주기억 장치 색인 기법의 특성을 고려하여, 본 논문에서는 Tmr-트리라는 새로운 색인 기법을 제시한다. Tmr-트리는 T-트리의 장점과 R-트리의 장점을 결합한 이진 색인 구조로서, 색인 노드는 데이터 객체들을 위한 엔트리들, 왼쪽/오른쪽 자식 노드에 대한 포인터, 그리고 3개의 추가 필드들로 구성된다. 여기서 3개의 추가 필드들은 현재 노드에 저장된 키 값들의 범위를 포함하는 MBR과 왼쪽 서브트리에 저장된 키 값들의 범위를 포함하는 MBR, 오른쪽 서브트리에 저장된 키 값들의 범위를 포함하는 MBR에 해당한다. 본 논문의 실험에서 Tmr-트리는 R-트리와 달리 검색 시 항상 리프노드까지 방문할 필요가 없기 때문에 모든 데이터 분포에서 R-트리에 비해 더 나은 실험 결과를 보여주었다. 노드 크기 측면에서 노드안 엔트리 수를 증가시킨 초반에 상당한 검색성능 향상을 보여주었으며, 그 후로 약간씩 검색시간 증가를 나타냈다. 한편, 삽입시간 측면에서 Tmr-트리는 R-트리에 비해 약간의 더 많은 삽입시간이 필요했다. As random access memory chip gets cheaper, it becomes affordable to realize main memory-based database systems. The disk-based spatial indexing techniques, however, cannot direct apply to main memory databases, because the main purpose of disk-based techniques is to reduce the number of disk accesses. In main memory-based indexing techniques, the node access time is much faster than that in disk-based indexing techniques, because all index nodes reside in a main memory. Unlike disk-based index techniques, main memory-based spatial indexing techniques must reduce key comparing time as well as node access time. In this paper, we propose an efficient spatial index structure for main memory-based databases, called Tmr-tree. Tmr-tree integrates the characteristics of R-tree and T-tree. Therefore, Nodes of Tmr-tree consist of several entries for data objects, main memory pointers to left and right child, and three additional fields. First is a MBR of a self node, which tightly encloses all data MBRs (Minimum Bounding Rectangles) in a current node, and second and third are MBRs of left and right sub-tree, respectively. Because Tmr-tree needs not to visit all leaf nodes, in terms of search time, proposed Tmr-tree outperforms R-tree in our experiments. As node size is increased, search time is drastically decreased followed by a gradual increase. However, in terms of insertion time, the performance of Tmr-tree was slightly lower than R-tree.

An efficient crash recovery technique for real-time main memory database

This paper presents an efficient recovery scheme suitable for real-time main memory database. In the recovery scheme, log records are stored in non-volatile RAM which is divided into four different partitions based on transaction types. Similarly, a main memory database is divided into four partitions based data types. When the using ratio of log store area exceeds the threshold value, checkpoint procedure is triggered. During executing checkpoint procedure, some useless log records are deleted. During restart recovery after a crash, partition reloading policy is adopted to assure that critical data are reloaded and restored in advance, so that the database system can be brought up before the entire database is reloaded into main memory. Therefore down time is obviously reduced. Simulation experiments show our recovery scheme obviously improves the system performance, and does a favor to meet the deadlines of real-time transactions.

An efficient cache conscious multi-dimensional index structure

Recently, to relieve the performance degradation caused by the bottleneck between CPU and main memory, cache conscious multi-dimensional index structures have been proposed. The ultimate goal of them is to reduce the space for entries so as to widen index trees and minimize the number of cache misses. The existing index structures can be classified into two approaches according to their entry reduction methods. One approach is to compress MBR keys by quantizing coordinate values to the fixed number of bits. The other approach is to store only the sides of minimum bounding regions (MBRs) that are different from their parents partially. The second approach works well when the size of a node is small and the number of entries is small. In this paper, we investigate the existing multi-dimensional index structures for main memory database systems through experiments under the various work loads. Then, we propose a new index structure that exploits the properties of the both techniques. We implement existing multi-dimensional index structures and the proposed index structure. We perform various experiments to show that our approach outperforms others.

실시간 모바일 GIS 응용 구축을 위한 주기억장치 데이터베이스 시스템 설계 및 구현

최근 들어 계속되는 램 가격 하락으로 인해 대용량의 램을 사용하는 주기억장치 데이터베이스 시스템의 구축이 실현 가능하게 되었다. 주기억장치 데이터베이스는 여러 다양한 실시간 응용 분야를 위해 사용되며, 매년 CPU 속도가 60％ 정도 증가되고, 메모리 속도가 10% 증가되는 현실에서, 케쉬 미스(Cache miss)를 얼마나 줄이느냐 하는 문제가 주기억장치 데이터베이스의 검색 성능 측면에서 가장 중요한 문제로 대두되고 있다. 본 논문에서는 이러한 환경을 고려한 실시간 모바일 GIS응용을 위한 주기억장치 데이터베이스 시스템을 설계 및 구현한다. 본 시스템은 크게 PDA를 사용하는 모바일 사용자를 위한 인터페이스 관리기와 가상 메모리 기법을 사용해 전체 데이터를 주기억장치에 상주시키며 관리하는 주기억 데이터 관리기, 공간 및 비 공간 질의를 처리하는 질의처리기, 새롭게 제시하는 공간 데이터를 위한 MR-트리 인덱스와 비 공간 데이터를 위한 T-트리 인덱스 구조를 관리하는 인덱스 관리기, 데이터를 디스크에 저장하기 위한 GIS 서버 인터페이스로 구성된다. 새롭게 제시하는 공간 인덱싱을 위한 MR-트리는 노트 분할이 발생될 경우, 입력 경로 상에 하나 이상의 빈 엔트리를 지니는 노드가 존재할 경우에만, 노드 분할을 상위로 전송한다. 그러므로 중간 노드들은 항상 100%에 가깝게 채워져 있게 된다. 본 논문의 실험 결과, 2차원의 MR-트리는 기존의 R-트리에 비해 2.4배 이상의 빠른 검색 속도를 나타냈다. 한편, 주 기억 데이터 관리기는 가상 메모리 제공을 위해 전체 벡터 데이터 및 MR-트리, T-트리, 데이터 객체 텍스트 정보를 페이지 단위로 분할하여 관리하고, 간접 주소 기법을 사용하여 디스크로부터의 재 로딩시 발생할 수 있는 문제점을 제거하였다. As random access memory chip gets cheaper, it becomes affordable to realize main memory-based database systems. Consequently, reducing cache misses emerges as the most important issue in current main memory databases, in which CPU speeds have been increasing at 60% per year, compared to the memory speeds at 10% per you. In this paper, we design and implement a main-memory database system for real-time mobile GIS. Our system is composed of 5 modules： the interface manager provides the interface for PDA users； the memory data manager controls spatial and non-spatial data in main-memory using virtual memory techniques； the query manager processes spatial and non-spatial query : the index manager manages the MR-tree index for spatial data and the T-tree index for non-spatial index : the GIS server interface provides the interface with disk-based GIS. The MR-tree proposed propagates node splits upward only if one of the internal nodes on the insertion path has empty space. Thus, the internal nodes of the MR-tree are almost 100% full. Our experimental study shows that the two-dimensional MR-tree performs search up to 2.4 times faster than the ordinary R-tree. To use virtual memory techniques, the memory data manager uses page tables for spatial data, non- spatial data, T-tree and MR-tree. And, it uses indirect addressing techniques for fast reloading from disk.

Effect of node size on the performance of cache-conscious B + -trees

In main-memory databases, the number of processor cache misses has a critical impact on the performance of the system. Cache-conscious indices are designed to improve performance by reducing the number of processor cache misses that are incurred during a search operation. Conventional wisdom suggests that the index's node size should be equal to the cache line size in order to minimize the number of cache misses and improve performance. As we show in this paper, this design choice ignores additional effects, such as the number of instructions executed and the number of TLB misses, which play a significant role in determining the overall performance. To capture the impact of node size on the performance of a cache-conscious B+ tree (CSB+-tree), we first develop an analytical model based on the fundamental components of the search process. This model is then validated with an actual implementation, demonstrating that the model is accurate. Both the analytical model and experiments confirm that using node sizes much larger than the cache line size can result in better search performance for the CSB+-tree.

고성능 고가용성을 위한 ALTIBASE(TM) 자료저장 관리기의 설계

기존 디스크 기반 데이터베이스 관리 디스트를 주저장 장치로 활용하는 특징적 환경 때문에 주기억 상주 데이터베이스 관리 시스템의 메모리 관리, 인덱스 관리, 자료저장 관리 기능 등에 대한 설계 및 구현 기술이 기본적으로 다르다. 본 논문에서는 현재 상용 시스템으로 사용되고 있는 <TEX>$ALTIBASE^{TM}$</TEX> 주기억 상주 DBMS 의 설계 및 구현 내용을 기술한다. 특히 자료저장 관리기의 세부 구성요소에 대한 구조적 특징가 주요 기능인 트랜잭션 처리 기법과 회복 관리 기법에 대하여 기존 DBMS 의 기능가 비교 설명하며, 고가용성을 위한 데이터베이스 이중화 기능 및 이중화 작업 시 발생하는 트랜잭션 충돌 해결 방법에 대하여 설명한다. 또한 <TEX>$ALTIBASE^{TM}$</TEX> 시스템의 성능을 측정하기 위하여 다양한 환경에서 실험된 TPS 결과를 보인다. Main memory database systems use the different implementation techniques to sturucture and organize the user dta and system catalogs, since traditional database systems are optimized for the characteristics of disk storage environment. We present, in this paper, the design considerations for our main memory database system <TEX>$ALTIBASE^{TM}$</TEX> that is currently applied to the time-critical applications. We focus on the design issues of storage manager in <TEX>$ALTIBASE^{TM}$</TEX>. The major components are introduced, and features and characteristics of transaction management and recovery method are described. We also present the database replication mechanism and its conflict resolution mechanism for high availability and performance. In order to evaluate our transaction performance, we show various experimental reports as measured by the TPS.

Affinity-based management of main memory database clusters

We study management strategies for main memory database clusters that are interposed between Internet applications and back-end databases as content caches. The task of management is to allocate data across individual cache databases and to route queries to the appropriate databases for execution. The goal is to maximize effective cache capacity and to minimize synchronization cost. We propose an affinity-based management system for main memory database cLUsters ( ALBUM ). ALBUM executes each query in two stages in order to take advantage of the query affinity that is observed in a wide range of applications. We evaluate the data/query distribution strategy in ALBUM with a set of trace-based simulations. The results show that ALBUM reduces cache miss ratio by a factor of 1.7 to 9 over alternative strategies. We have implemented a prototype of ALBUM, and compare its performance to that of an existing infrastructure: a fully replicated database with large buffer cache. The results show that ALBUM outperforms the existing infrastructure with the same number of server machines by a factor of 2 to 7, and that ALBUM with only 1/3 to 1/2 of the server machines achieves the same throughput as the existing infrastructure.

Sunrise: A real-time event-processing system

Conceived at Bell Labs, Sunrise is a high-throughput, real-time event-processing and aggregation system. It offers many of the data-management features of a conventional database system, such as high-level, declarative programming interfaces and the traditional atomicity, consistency, isolation, and durability (ACID) correctness guarantees for transactions. Adapted to meet the needs of high-performance applications, Sunrise maintains its high throughput by using a shared-nothing parallel architecture that can deliver close to linear scale-up in practice. Sunrise achieves the real-time responsiveness (on the order of as little as a few milliseconds per transaction) that many emerging applications require by relying on the DataBlitz™ main-memory database system as its underlying single-site storage manager. The DataBlitz storage manager guarantees the fast access to correct data expected from a database system. The strength of Sunrise lies not only in its performance, but also in its flexibility. This paper introduces Sunrise and its two major components: a real-time analysis engine (RAE), at the heart of the runtime system, and a service authoring environment (SAE), which is the Sunrise authoring interface.

3106 主記憶データベースが実現する高速分散協調設計環境の提案

3106 主記憶データベースが実現する高速分散協調設計環境の提案

Improving index performance through prefetching

This paper proposes and evaluate Prefetching B + - Trees (pB + -Trees), which use prefetching to accelerate two important operations on B + -Tree indices: searches and range scans. To accelerate searches, pB + -Trees use prefetching to effectively create wider nodes than the natural data transfer size: e.g., eight vs. one cache lines or disk pages. These wider nodes reduce the height of the B + -Tree, thereby decreasing the number of expensive misses when going from parent to child without significantly increasing the cost of fetching a given node. Our results show that this technique speeds up search and update times by a factor of 1.21-1.5 for main-memory B + -Trees. In addition, it outperforms and is complementary to “Cache-Sensitive B + -Trees.” To accelerate range scans, pB + -Trees provide arrays of pointers to their leaf nodes. These allow the pB + -Tree to prefetch arbitrarily far ahead, even for nonclustered indices, thereby hiding the normally expensive cache misses associated with traversing the leaves within the range. Our results show that this technique yields over a sixfold speedup on range scans of 1000+ keys. Although our experimental evaluation focuses on main memory databases, the techniques that we propose are also applicable to hiding disk latency.

Main-memory index structures with fixed-size partial keys

The performance of main-memory index structures is increasingly determined by the number of CPU cache misses incurred when traversing the index. When keys are stored indirectly, as is standard in main-memory databases, the cost of key retrieval in terms of cache misses can dominate the cost of an index traversal. Yet it is inefficient in both time and space to store even moderate sized keys directly in index nodes. In this paper, we investigate the performance of tree structures suitable for OLTP workloads in the face of expensive cache misses and non-trivial key sizes. We propose two index structures, pkT-trees and pkB-trees, which significantly reduce cache misses by storing partial-key information in the index. We show that a small, fixed amount of key information allows most cache misses to be avoided, allowing for a simple node structure and efficient implementation. Finally, we study the performance and cache behavior of partial-key trees by comparing them with other main-memory tree structures for a wide variety of key sizes and key value distributions.

AVL Trees with Relaxed Balance

The idea of relaxed balance is to uncouple the rebalancing in search trees from the updating in order to speed up request processing in main-memory databases. In this paper, we describe a relaxed version of AVL trees. We prove that each update gives rise to at most a logarithmic number of rebalancing operations and that the number of rebalancing operations in the semidynamic case is amortized constant.

Incorporating load factor into the scheduling of soft real-time transactions for main memory databases

Many real-time applications have very tight time constraints which couldn't be met by disk resident databases. For those applications, main memory database where entire database is stored in main memory is the proper choice. It has been shown that coarse-granule locking is better than fine-granule locking for main-memory databases. Coarse-granule locking makes it easy to extract data access patterns correctly from canned transactions of main memory real-time database systems. In this paper, we propose two real-time transaction scheduling algorithms — CCA-ALF (Cost Conscious Approach with Average Load Factor) and EDF-CR-ALF (Earliest Deadline First-Conditional Restart with ALF) — which use both static (e.g., deadline) and dynamic information (e.g., system load) for main memory databases by utilizing data access patterns of transactions. We compare the performance of those algorithms with CCA and EDF-HP which do not use system load information at all. Our simulations on main memory databases indicate that: i) CCA-ALF is better than EDF-HP, CCA, and EDF-CR-ALF in terms of miss percent and mean lateness, and ii) CCA-ALF adapts well to the changes in the system load.