VERITAS Database Edition for Sybase. Technical White Paper

Transcription

1 VERITAS Database Edition for Sybase Technical White Paper M A R C H

2 Introduction Data availability is a concern now more than ever, especially when it comes to having access to mission-critical databases. Not only do users require 24x7 access to data, they also want increasingly fast access to that data. The amount of data in databases has been increasing exponentially, causing a nightmare for database administrators who must manage that data to achieve the desired fast access. Due to this growth, finding experienced database administrators has become more difficult, making management of raw partitions a thing of the past. New administrators would much rather manage databases on top of file systems, and if a graphical interface is available even better. VERITAS Database Edition for Sybase meets the needs of database environments by incorporating the strengths of the VERITAS technology products with database-specific enhancements for servers running Sybase Adaptive Server Enterprise (ASE) 11.5 or later. VERITAS Database Edition for Sybase offers unrivaled availability, performance, and manageability for Sybase environments. The Database Edition not only combines the performance and data integrity of raw disk partitions with the manageability of file systems, but it also has been certified by Sybase. This paper will examine how VERITAS Database Edition for Sybase software enhances the availability, performance, and manageability of Sybase RDBMS, one of the most common systems for deployment of mission-critical applications today. VERITAS Database Edition for Sybase VERITAS Database Edition for Sybase is an integrated suite of storage and failover management software designed to provide enhanced availability, increased performance, and reduced administration cost for Sybase dataservers. The Database Edition includes the following VERITAS products: VERITAS File System (VxFS) VERITAS Volume Manager (VxVM) VERITAS Database Edition/HA for Sybase adds: VERITAS Cluster Server VERITAS Cluster Server Agent for Sybase VERITAS File System VERITAS File System is an extent-based, journaling file system, for increased performance and accelerated recovery from system failure. Its online administration support allows file systems to be resized, defragmented, and backed up without interrupting applications or users. Its design is optimized for large files and large I/O operations, which is especially useful for integration with dataserver export operations and for large data items (video, images, audio) stored external to (and pointed to by) a dataserver. VERITAS file system interface technology, VERITAS Quick I/O, enhances Sybase dataserver performance. Quick I/O presents a pre-allocated VERITAS File System file to dataservers as a raw character device, enabling them to achieve raw device performance, especially for online transaction processing (OLTP) and write-intensive workloads. Database Edition for Sybase also offers Cached Quick I/O, an enhanced feature of Quick I/O, which enables databases on systems with large available memory to take advantage of the additional memory to improve performance. VERITAS Volume Manager VERITAS Volume Manager is a comprehensive disk storage management system, providing online configuration, administration, and tuning tools. It incorporates disk spanning and RAID capabilities including striping (RAID-0), mirroring (RAID-1), mirrored striping (RAID-0+1), striped mirroring (RAID-1+0) and striping with distributed parity (RAID-5), to enhance the reliability and performance of server storage. VERITAS Volume Manager also features a Java-based, easy-to-use graphical interface, VERITAS Storage Administrator, to simplify file system and database storage management. When used with Sybase, VERITAS Volume Manager can increase the throughput of multi-user access by distributing I/O requests across multiple spindles. Additionally, mirroring and RAID-5 keep dataserver data available in the event of disk failure. The online configuration capabilities allow mirroring and striping characteristics, size, and location of data to be changed without interrupting transaction access. Use of VERITAS Volume Manager is transparent to Sybase, which accesses a volume as if it were a single disk partition, enhancing the performance and online management capabilities of the dataserver. VERITAS Storage Administrator provides a Java-based graphical user interface to VERITAS Volume Manager and VERITAS File System. It also provides visual elements such as icons, menus, and forms to ease the task of manipulating Volume Manager objects and file systems. File systems and volumes created using Storage Administrator are fully interoperable with those created via the command line interface, which can be used to create scripts for dataserver administration. 1

3 VERITAS Cluster Server VERITAS Cluster Server provides flexible, easy-to-configure application availability management, allowing file and database client/server configurations to survive any single point of failure, as well as provide the ability to schedule service tasks with little to no interruption to users. VERITAS Cluster Server enables automatic recovery of the dataserver and application services by transferring workloads to one of up to 32 servers in the cluster. For simplified integration, agent support for Sybase is incorporated. The VERITAS Cluster Server Agent for Sybase detects Sybase instance failure; it tries to restart the dataserver on the same system if possible, or initiates failover to another system in the cluster to keep databases accessible. Combining these online storage and availability management tools, the Database Edition for Sybase delivers unrivaled performance, high availability, and enhanced manageability to Sybase dataservers. Users Network VERITAS Database Edition for Sybase & HA Edition End-User Applications Storage Administrator VCS+ Agent File System + (Cached) Quick I/O VERITAS Volume Manager Applications Sybase Server Operating System Disks Running Sybase Dataservers on VERITAS File System Database tables and other structures may be stored in two different types of storage containers: 1) they may use raw disk partitions ( raw files ) that avoid the file system layer entirely or 2) they may be configured in standard operating system files ( cooked files ). While raw disk allocation potentially provides increased performance and data integrity compared to standard UFS file systems, these considerations are not meaningful for VERITAS File System. VERITAS File System allows users to attain equal levels of performance and data integrity as are available with raw partitions while using a consistent naming convention for files that provides system administrators with an added level of comfort and manageability. Building a dataserver on VERITAS File System allows administrators to use common tools and strategies for backing up and restoring dataservers and other system files. Thus, VERITAS File System combines the performance of raw disk with the administrative benefits of traditional file systems. Data Integrity and Performance Issues with File Systems vs. Raw Disk Partitions Sybase strongly recommended its users to configure database devices using raw disk partitions. Sybase has traditionally advocated raw partitions to address data integrity and performance concerns normally associated with storing databases in files. DBMS using raw disk partitions rely on the fact that writes to raw disk are performed immediately. These products use their own buffer pools in order to manage page allocation more efficiently for the DBMS than the standard file system cache manager does. The additional buffering imposed by the traditional file system is both an inefficient use of memory and potentially dangerous. For example, if the DBMS pager believes that blocks have been flushed to disk when they are really sitting in the operating system s buffer cache, data could be lost in the event of system failure. While raw disk partitions typically provide higher levels of data integrity and performance, they are less flexible to administer than file systems. Raw partitions are not visible, which can easily lead to operator error. In contrast, by building dataservers on file systems, administrators can store database tables in identifiable locations. Backups of raw partitions are limited to using 2

4 the dd command or highly specialized and invasive backup software like SQL-BackTrack. Whereas, file systems can be backed up using third party products that use the Sybase Backup API such as VERITAS NetBackup for Sybase. In addition, dataserver growth is more difficult to manage with raw partitions. VERITAS File System allows the DBA greater flexibility in allocating new database devices for creating new segments or extending existing segments. Sybase has certified its dataservers to run on top of Database Edition for Sybase. Working with the Quick I/O file device driver (VRTSqio), VERITAS File System allows a regular, pre-allocated file to be accessed as a raw character device. This not only eliminates any data integrity concerns, but also achieves the same dataserver performance as raw devices. VERITAS File System Reliability and Performance Optimization VERITAS File System addresses performance and reliability issues for databases stored in file system files, providing the administrative benefits of file systems without loss of performance or data integrity. Unlike UFS, which uses block-based allocation schemes, VERITAS File System is extent-based. The extent-based design provides considerably better performance for sequential I/O operations such as table scans and for large file I/Os (such as those for multimedia applications). Furthermore, VERITAS File System addresses the aforementioned data integrity and performance issues by using Quick I/O. VERITAS Quick I/O Database Accelerator Using VERITAS Quick I/O, dataservers running on VERITAS File System deliver performance equal to those running on raw partitions. Quick I/O allows a regular VERITAS File System file to be accessed by the DBMS software as a raw partition. Quick I/O bypasses the UNIX buffer cache, eliminating extra buffering. This technology ensures data integrity by writing data safely from the dataserver s own buffer cache to the disk. Without extra buffering, the dataserver also eliminates redundant buffer copying, and thus excess data copying on every read and write. In addition, by avoiding the additional buffering, Quick I/O allows for more efficient use of memory and significantly reduces CPU overhead. Quick I/O attains its performance gains through shared update locks and asynchronous I/O. Quick I/O supports shared update locks for database files at the VERITAS File System level. Traditionally, the exclusive write lock at the UNIX file system level has caused write transactions to be serialized, slowing down database performance, especially with many simultaneous users. Quick I/O eliminates this bottleneck and allows the DBMS to process parallel update requests. Quick I/O further improves dataserver performance by taking advantage of operating system supported asynchronous I/O, which has previously been available only to raw partitions. For example, Quick I/O now takes advantage of Solaris enhanced kernelized asynchronous I/O (KAIO) for dataservers built on VERITAS File System. Cached Quick I/O, an enhanced feature of VERITAS File System s Quick I/O technology, enables databases on systems with large available memory to take advantage of the additional memory to improve performance. The use of Cached Quick I/O has enabled performance increases over that of raw devices for Sybase databases in environments with a high percentage of reads. For read operations, Cached Quick I/O allows database blocks to be cached in the system buffer cache. This caching potentially reduces the number of physical I/O operations and thus improves the read performance. For write operations, Cached Quick I/O uses a direct-write, copy-behind technique to preserve its buffer copy of the data. After the direct I/O is scheduled, and while it is waiting for the completion of the I/O, the file system updates its buffer to reflect the changed data being written out. Therefore, for online transaction processing, Cached Quick I/O achieves better than raw performance for systems with large physical memory without having to manage a large buffer cache. Data and System Availability Issues As businesses increasingly depend on DBMS for managing mission-critical data, the requirement for data and dataserver availability also increases. VERITAS File System improves data availability with its online administration and fast recovery features. These features reduce planned and unplanned downtime for business critical dataservers. Online Administration of Dataserver Storage Configurations VERITAS File System s online design allows administrators to defragment or resize a file system while it remains accessible to dataservers and their users. A file system can be expanded online only when the underlying device can be expanded online. The Database Edition provides online file system growth capability just as VERITAS Volume Manager allows online expandability of virtual disks. The online administration capabilities simplify the dataserver maintenance task of adding database devices and segments to meet growing storage requirements. Fast File System Recovery to Minimize Dataserver Downtime VERITAS File System is a journaling (intent-logging) file system, providing fast recovery. VERITAS File System requires only seconds to recover after a system crash or reboot with full system integrity, alleviating the long recovery problem associated with traditional file systems. Similar to dataserver recovery, VERITAS File System recovery requires an intent-log replay, committing or aborting any pending file system operations. This technique allows data on file systems to be accessible to dataservers immediately after the log replay, minimizing dataserver downtime. 3

5 Sybase Backup Issues Sybase users generally use the built-in Backup Server for online backup, recovery, and archiving, using the dump and load commands. Third party backup products such as VERITAS NetBackup work with Sybase Backup Server APIs to provide automated backup in networked environments and manage backup media including robotic devices. This method works well for users who want to recover granular data at the database and transaction levels. However, it requires detailed knowledge of the physical and logical database layouts. Some users may prefer to shutdown the dataserver and do a cold backup of all database devices. Cold backup is becoming less and less practical in this 24x7 uptime world due to the long period the dataserver must be taken offline to perform the backup. In accordance, Sybase added the ability to quiesce its database to perform hot backup with the introduction of ASE VERITAS File System Snapshot to Minimize Dataserver Backup Window Once the file system is used for dataserver storage, the backup strategy can utilize a VERITAS File System feature, snapshot. The VERITAS File System snapshot mechanism allows the administrator to configure a small amount of additional storage (generally as little as 5% of the storage used for data) to present a consistent view of the data frozen at a single point in time. When used in conjunction with dataserver backup strategies, the snapshot mechanism reduces backup windows to a couple of seconds and simplifies restore procedures. Although cold backups are easy to perform, provide a simple restore procedure, and safeguard against media failures, it is costly to shutdown mission-critical dataservers. The time required to snapshot a file system is typically only a few seconds; therefore, the administrator can obtain a fully consistent image of a database or an entire system with virtually no dataserver downtime. After shutting down a dataserver, the administrator can take a snapshot of all file systems containing database files. The dataserver can then be brought back online to resume operations. At the same time, the administrator can backup from the snapshot. With the introduction of ASE 12.0, Sybase added the ability to quiesce the database, so there is no longer the need to shutdown the database to perform backup. When using this hot technique, Sybase suspends all updates to the database until the database is no longer quiesced. Adding the VERITAS File System snapshot to this technique minimizes the amount of time the database is quiesced from processing updates, thereby reducing down time due to backup. Use of VERITAS File System with Quick I/O in Sybase environments allows users to take advantage of the benefits of raw disk integrity and performance, coupled with the manageability offered by file systems. VERITAS File System also increases dataserver availability with its online administration features and reduces unplanned downtime by recovering from system failures in only a few seconds. Additionally, the VERITAS File System snapshot mechanism shortens the backup window when performing dataserver backup, dramatically reducing costly downtime associated with mission critical dataservers. Using VERITAS Volume Manager with Sybase Most database applications require higher availability and performance levels than those that can normally be attained using standalone disk drives. VERITAS Volume Manager enhances the storage management capabilities of Sybase by providing the enterprise-class disk management capabilities needed in large-scale, multi-disk environments. When used with Sybase, VERITAS Volume Manager provides the performance and reliability enhancements offered by hardware RAID storage implementations combined with the flexibility of a software tool. VERITAS Volume Manager may be configured with standalone disk drives; alternatively, it may be used to enhance controller-based RAID solutions. It incorporates not only tools to monitor and tune performance, but also a simple and unified graphical interface, VERITAS Storage Administrator, for dataserver storage configuration. The VERITAS Volume Manager management tasks may be performed while users are online, without loss of data or availability, and without requiring costly data backup/reload cycles for many reconfiguration operations. Increasing Performance and Data Integrity with VERITAS Volume Manager With VERITAS Volume Manager, administrators can enhance the performance and integrity of dataservers and simplify their storage administration using standard disk drives and controllers as well as special-purpose hardware-based RAID subsystems. Dataserver Storage Configurations for Enhanced Performance and Availability VERITAS Volume Manager offers various dataserver storage layout options, allowing administrators to optimize performance and to satisfy the availability requirements of specific database applications. These storage layouts include: Striping (RAID-0), for increased performance for multi-user read and write-intensive workloads. Striping distributes the I/O access for single segments across multiple disks, allowing parallel data transfer. 4

6 Mirroring (RAID-1), for high data availability in a high-performance environment. By allocating space for multiple copies of database storage, administrators can attain high availability and high performance for write- and read-intensive workloads, allowing users to continue accessing data reliably in the event of disk failure. Mirrored striping (RAID-0+1), combining the performance benefits of striping with the availability benefits of mirroring. Striped mirroring (RAID-1+0), combining the performance benefits of striping with the availability benefits of mirroring with the added benefit of enhanced redundancy for faster recovery from media failure. Parity striping (RAID-5) allows administrators to configure highly available storage using fewer disks than required by mirrored solutions. While RAID-5 configurations do not perform as well as mirrored stripes in write-intensive configurations or in the event of disk failure, they provide a cost-effective solution for query-intensive applications, including many data warehousing applications. Spanning (concatenation), for sequential access to database tables or logs by creating a contiguous virtual disk device. Spanning is useful for reading from or writing to database transaction logs when there is no sufficient contiguous physical space available. VERITAS Volume Manager s mirroring and RAID-5 capabilities protect data from subsystem failures, increasing data availability. Its hot relocation feature provides an additional level of protection by automatically restoring data redundancy in mirrored and RAID-5 configurations when a disk fails. It accomplishes this by relocating the affected portions to disks designated as spares and/or free space within the disk group, which minimizes vulnerability to a subsequent failure. Using VERITAS Volume Manager and Hardware RAID Because VERITAS Volume Manager is host-based, it allows the administrator far more flexibility than that offered by most hardware-only solutions. For example, VERITAS Volume Manager allows administrators to use portions of disk subsystems, and even of single drives, for different purposes in different configurations. Many hardware solutions do not provide allocation granularity smaller than single disk drives, or in some cases, entire controllers. However, administrators may choose to combine VERITAS Volume Manager s software RAID capabilities with hardware arrays, merging the controller s RAID-5 performance with the flexibility of VERITAS Volume Manager. Uses include: Mirroring hardware RAID configurations, removing the RAID controller as a single point of failure and enhancing read performance. Partitioning hardware arrays using VERITAS Volume Manager. When hardware-only RAID subsystems are configured using large logical units (LUNs), limitations in available partition tables and usable database devices may result in wasted space. VERITAS Volume Manager has a flexible virtual partition table, eliminating this problem. VERITAS Volume Manager s Flexible Administration The VERITAS Volume Manager online administration model assists administrators in allocating storage and managing availability and performance without interrupting user access. VERITAS Volume Manager enhances dataserver storage administration in several ways including: VERITAS Volume Manager online reconfiguration feature allows administrators to add mirrors to a dataserver, remove mirrors, or offline a mirror for stable backup and then return it to online status without interrupting operation or reconfiguring database tables. While this approach to backup (only available for ASE 12.0) requires more storage than the VERITAS File System snapshot capability described previously, it may be used with raw volume devices. The VERITAS Volume Manager configuration is managed at the device level rather than through the dataserver schema, simplifying storage management. Online storage relayout allows for the changes in the storage layout to increase performance. For instance, stripe depth and width or RAID level can be altered while users are accessing data to boost database application performance. Online resizing of storage area (both the virtual disk device and the file system created on the device) to support changing dataserver storage requirements. A Java-based, easy-to-use graphical user interface, Storage Administrator, in addition to the command line interface, simplifies file system and disk storage management. 5

7 The RAID capabilities of VERITAS Volume Manager increase the availability and performance of Sybase dataservers by offering various dataserver storage layouts to meet specific database applications such as OLTP or data warehousing. The online management features allow administrators to reconfigure dataserver storage and optimize performance without interrupting users. Using VERITAS Cluster Server with Sybase -- (Database Edition/HA for Sybase) VERITAS Cluster Server is a high availability application management solution for network services in the client/server environment. By offering monitoring and recovery functionality for applications, dataservers, network interfaces, and other components, VERITAS Cluster Server goes beyond machine-to-machine failover to provide a comprehensive high availability solution for any application service. Designed with a modular, extensible architecture to make it easy to install, configure and modify, VERITAS Cluster Server fully automates application-level fault detection, isolation, and recovery. Monitored service groups and resources can either be restarted locally or migrated to another node and restarted. A service group may include an unlimited number of resources. Used with a Sybase-specific agent, VERITAS Cluster Server monitors the dataserver processes as well as the system resources to recover automatically and quickly from DBMS software failure. When a system fails or the VERITAS Cluster Server Agent for Sybase detects a DBMS failure, VERITAS Cluster Server fails over the dataserver services to one of up to 32 servers in the cluster. To support DBMS failover, systems are configured so that the SAN or shared SCSI disks contain the dataserver data and software. These disks should have the same mount points and device names on both machines. When a failover occurs, the dataserver service is restarted on the takeover server, using the dataserver data and executables on the shared disk. In addition, the Agent Developer s Kit (ADK) allows developers to extend the high availability environment to manage failover of application and network services that are unique to the site. VERITAS Cluster Server s ease of configuration, through centralized configuration files and simple installation templates, makes it simple for DBAs and system administrators to configure a complete high-availability solution for database applications. With its simple and non-intrusive design, VERITAS Cluster Server enables administrators to build cost-effective, reliable, high availability cluster configurations for Sybase servers. Combined with a Sybase-specific agent, VERITAS Cluster Server significantly increases server and dataserver availability, ensuring continuous access to business critical data in client/server configurations. Tuning Guidelines for Sybase Dataservers The Database Edition for Sybase offers the best storage management solution for dataservers, which maximizes performance with minimal overhead. The following are some performance tuning guidelines that may be helpful when using VERITAS Database Edition to configure a Sybase dataserver. 6

8 Using Striped Volumes for Performance In general, the best way to achieve the optimal OLTP throughput is by balancing the I/O across as many disks as possible. This throughput can be achieved by creating the dataserver on a file system on a striped (RAID-0) volume using all of the available disks. However, since the dataserver logs tend to be accessed sequentially, one may see slightly better performance by putting the dataserver logs on a separate file system using a simple volume on a dedicated disk. If dataserver integrity is important, as it usually is, mirror (using RAID-1) at least the volume used by the dataserver logs. VERITAS Database Edition s support for disk mirroring, reconfiguration, and replacement allows online system administration and dynamic performance tuning without any downtime in a production dataserver. Using Quick I/O Technology The default UFS supports buffered I/O, in which data blocks are buffered in the operating system kernel cache. Buffered I/O often speeds up repeated reads on the same data blocks. This buffering is useful for smaller files that are accessed frequently. However, since the dataserver is maintaining its own cache, buffering the data in the kernel actually increases CPU overhead because of data copying between Sybase cache and kernel cache. In addition, UNIX semantics require that only one writer have the exclusive right to update a file, which forces other read and write operations on the same file to wait. As a result, a dataserver on file systems often performs slower than a dataserver on raw devices. Quick I/O has I/O characteristics similar to those of raw I/O, which have a higher degree of read/write parallelism with less CPU overhead. Quick I/O sustains higher transaction throughput than buffered I/O in an OLTP environment. Tuning the Dataserver Buffer Cache With the Database Edition, the administrator no longer needs to configure enough memory for the file system buffer cache and balance the memory usage between its cache and Sybase s cache. When Quick I/O is used, the dataserver cache should be tuned as though raw partitions are used. This way more memory can be allocated for the dataserver buffer cache, which often improves OLTP performance significantly. Summary VERITAS first made the Database Edition for Sybase available in October 1998 after receiving certification from Sybase. The fifth in its family of application-specific storage and failover management suites, the VERITAS Database Edition for Sybase is optimized for the Sybase environment. It combines the performance and data integrity of raw devices with the manageability of building Sybase dataservers on top of file systems. VERITAS Database Edition for Sybase can now be used to enhance Sybase dataservers in the following ways: Performance: Dataserver performance equal to raw device levels using VERITAS File System with Quick I/O. Cached Quick I/O for better than raw performance for systems with large available memory and a high percentage of read operations. Excellent performance for large files such as images, audio, and video using VERITAS File System s extent-based allocation. Availability: The highest level of dataserver and file system availability in the event of downtime caused by system failure or scheduled maintenance, using VERITAS Cluster Server and VERITAS Cluster Server Agent for Sybase. RAID redundancy techniques to prevent data loss caused by media failure. Hot relocation to minimize vulnerability to subsequent media failure by restoring redundancy. Online administration for storage reconfiguration, storage relayout, and maintenance tasks. Manageability: Storage Administrator eases management tasks with its Java-based GUI for Volume Manager and File System. Wide choice of configuration options for optimal balance of performance, availability, and cost for different types of data for example, use of RAID-5 volumes for read-only historical tables. Simple installation and configuration of VERITAS Cluster Server to allow quick, reliable deployment of high availability services to make sure data is safe and secure. 7

9 Corporate Headquarters 1600 Plymouth Street Mountain View, CA North American Sales Headquarters 400 International Parkway Heathrow, FL or Fax Global Locations United Kingdom or 44-(0) (0) Fax France Fax Germany 49-(0) (0) Fax South Africa Fax Australia BACKUP 61-(0) Fax Hong Kong Fax Japan Fax Malaysia Fax Singapore Fax China Fax Electronic communication World Wide Web: VERITAS Software Corporation. All rights reserved. VERITAS is a registered trademark of VERITAS Software Corporation in the US and other countries. The VERITAS logo, Business Without Interruption, VERITAS Database Edition, VERITAS Cluster Server, VERITAS File System and VERITAS Volume Manager are trademarks of VERITAS Software Corporation in the US and other countries. Other product names mentioned herein may be trademarks and/or registered trademarks of their respective companies. Specifications and product offerings subject to change without notice. Printed in USA. March xxx-910 8