1295-1297

Page 1295

Another significant direction in storage technology is the availability of storage subsystems that present a single device appearance to the operating system, but internally use many disks and gigabytes of cache memory to hide device boundaries and mask failures. With a single logical device capable of storing hundreds of gigabytes and transfer rates of hundreds of megabytes per second, the pure shared-nothing limitation of access to the device by one instance only is a severe compromise in processing power. With Oracle's unrestricted access implementation, all the processors in the system can access all the disks in the logical device without going through a "master" instance, delivering scalable performance.

Parallel Processing Platform Hardware Configuration Requirements

In this section, I discuss a few aspects of the required hardware configuration for a parallel processing platform.

Fail-Safe Redundant Hosts

HP, Sun, and Digital 64-bit UNIX provide for an instantaneous redundant host. In the event of a system-wide failure, a "switchover" of the production UNIX servers disks and software can be transferred to a development host automatically. The HP Switch Over daemon sends a regularly scheduled signal or "heartbeat" and receives state-of-health diagnostic information in response to that signal, or absence of a signal to the standby host.

When the standby host determines that the messages have stopped , users are warned of a pending shutdown. HP Switch Over then initiates a takeover, locking the current host's disks, and rebooting as the current host. Heartbeat sends messages to notify the dead or distressed host that a switchover sequence has occurred and issues the appropriate operating system shutdown, ensuring that the lame machine notices the takeover and halts.

The DEC Alpha file servers 2100, 8200, and 8400, utilizing the model EV/5 processor deliver added redundancy through "CPU failover" and "memory failover." If a single CPU or RAM failure occurs during processing, the SMP file server continues uninterrupted processing! At the time of failure, a "system in distress" page is transmitted to service support personnel for immediate notification of the particular problem.

Cross or Remote Mounted Disk Drives

Remote storage devices should not be used in parallel processing because they must interface with at least the two (2) file servers. Any read/write activity is effectively doubled , between the disks of a remote mounted file system as they must interface with each disk's separate processor in the configuration.

Remote file systems must also interface with the network router, cable, connectors, concentrators , Adaptable User Interfaces (AUIs), host adapters, smart controller cards, and buses in the

Page 1296

path connecting these two systems. Remote file systems effectively double seek time at a minimum, but they can also degrade network performance for the entire Local Area Networks involved.

Disk Drive Allocation

The number of physical disks mounted and available to the file system are critical to the performance of Oracle and related client/server applications. Performance bottlenecks can occur because there is access to only a single "RAID5 stripe" or a single input/output disk read/write "needle," regardless of the number of logical volumes per physical disk drive.

The physical and not logical volume devices are used to balance the I/O of the background Oracle processes across the disks.

Disk Partitioning

Physical disks may be "partitioned" or configured by a logical device manager in subunits that may allow a single disk drive to appear to be multiple disk drives, with different device names . The converse is also true as several disks may appear to be a single disk partition.

Before you install Oracle, check with your operating system administrator for clarification of these physical, not the logical, device names. Also verify the necessary path, permissions, and storage space of the data files that you can create on each disk drive.

Maximizing Parallel Platform Database and Disk Performance
Through Balancing the Oracle Processes

The system tablespace, rollback segments, control files, and redo logs are always active for every Oracle instance. If Archive mode is enabled, then transaction logging will "journal" all data-base activity enabling a point-in-time recovery if needed, in conjunction with your full database backup. Archive mode should write to a duplexed tape and disk media. The fastest recovery will be directly from disk, if this is possible.

These processes use specific files, which should be distributed across different disk drives for optimal database speed.

Be certain that all Oracle background and foreground processes have the same operating system process priority. Contention will arise as a high-priority process waits for a low-priority process, which may never swap back in.

Disk Optimization for Oracle and the Parallel Server Option

The optimum number of disks for the Oracle installation should at a minimum enable the separation of the table data on disk one and the index data on disk two, in any size .

With additional disk availability, the Oracle system offers tablespace on disk three, rollback segments on disk four, temporary tablespace on disk five, and the redo logs on disk six.

Page 1297

The triple- copied control files are always written to when major RDBMS events occur and are not a serious performance bottleneck. The control files should always be separated across multiple disks, whenever more than one disk drive is available by direct mount to the file system. This triple redundancy ensures rapid recovery in the event of a disk drive failure or accidental erasure.

Time zones and business application groups should influence the tablespace design so that a database maintenance window for one application does not interfere with the operation of 7/24 maximum availability systems.

Disk Storage Devices

The Oracle disk storage devices in Table 56.2 can be standard devices, mirrored devices, Write Once Read Many optical disc (WORM), Write Many Read Many optical disc (WMRM), Redundant Arrays of Independent Disks (RAID) and other media types. These disks should have relatively high read/write rate (8_9 milliseconds for standard disk drives) and should not be cross-mounted from other file system servers, except in the case of an out-of-file-space emergency.

Optical Storage

The media read and write rates for optical storage are significantly slower and therefore should not be used as high-performance file system devices for normal database query activity. These optical devices are very well suited for backups , long- term data archival, and are usually guaranteed for long-term readability of 50 years or more.

Table 56.2. Storage device pricing and performance.

continues

Similar products