The Write Accelerator is specifically designed to increase Oracle database performance. It accelerates Oracle databases by allowing them to handle substantially greater transaction volumes and add more concurrent users. The Write Accelerator can deliver many times the performance of a standard cached-RAID array and is more easily installed and managed by enterprises whose operations rely on Oracle-resident applications.
The Write Accelerator works on a simple premise: performance of write intensive databases is limited by two factors – latency and throughput. The Write Accelerator, a special edition RamSan solid state disk solution, offers unparalleled latency and throughput and unbinds Oracle applications limited by slow writes.
Solutions designed to accelerate writes must meet three requirements: tightly integrate with the operating systems that run Oracle, support Oracle and Oracle ASM, and follow Oracle’s stripe and mirror everything (SAME) rule. The Write Accelerator supports Oracle in Windows, Linux, Solaris, HP-UX, OpenVMS, and AIX environments, among others. The systems have been deployed with Oracle databases from Oracle 8i to Oracle 11g, including support for Oracle RAC and Oracle ASM. The Write Accelerator consists of two separate but bundled units, enabling complete support for SAME and allowing even the most demanding enterprises in the world 5-nines reliability for the solution.
The Write Accelerator can be directly attached to Oracle database servers or attached to storage networking components. Special skills or training are not required to install and administer a Write Accelerator. Administrators simply move their redo logs, undo segments, or temporary tablespaces to the Write Accelerator. Neither the Oracle database itself nor the operating system need to be modified to support the Write Accelerator.
“Write activity in Oracle is mostly a factor of the performance of the redo logs, undo segments, and temporary tablespaces,” says Mike Ault, Oracle Guru at Texas Memory Systems and author of more than 20 books on Oracle performance and tuning-related topics. “By moving these files to the Write Accelerator, application performance is boosted as server resources are made available to support more transactions and more concurrent users. Plugging it in is easier, more effective, and less costly than adding additional RAID arrays or constantly tuning the database.”
Oracle database administrators can determine if their application suffers from write contention or other I/O performance issues by analyzing their Oracle Statspack and Automated Workload Repository (AWR) reports using the free tool at StatspackAnalyzer.com, an Oracle community forum co-sponsored by Texas Memory Systems and Burleson Consulting.
The Write Accelerator connects to servers or network fabrics through basic Fibre Channel or InfiniBand interfaces and appears simply as a very fast disk to the network, able to act as any disk or storage system might. The Write Accelerator acheives dramatic performance gains in Oracle redo logs, temporary tablespaces, and undo segments.
The Write Accelerator solution incorporates two individual solid state disk systems mirrored in host-side or Oracle ASM configurations, providing the highest degree of data protection and system reliability. In addition, each individual unit of the Write Accelerator solution includes multiple reliability features, including hot swappable power supplies and backup hard disks, redundant cooling fans, batteries, and optional redundant failover data ports. ChipKill technology, ECC and soft error scrubbing, and Active Backup, as noted below, are also part of the Write Accelerator’s suite of sophisticated reliability features.
The Write Accelerator is designed to look like a disk to the network. It is highly interoperable and works in virtually any enterprise environment. It is installed in numerous environments, including AIX, Solaris, HP-UX, OpenVMS, Tandem, Linux, and Windows. Hardware from Fibre Channel and InfiniBand vendors including Cisco, Emulex, Brocade, McData, Mellanox, QLogic, LSI Logic, and ATTO easily interoperates with the Write Accelerator.
The Write Accelerator includes the reliability and fuctionality of Active Backup®. Other SSDs and cache systems can only "destage" data from RAM to backup disks once power is already lost. This means that data is never backed up during normal operation and the time-to-complete-backup after power loss is unacceptably long. A Write Accelerator with Active Backup can continuously back up data on RAM to the internal backup hard disks without impacting performance. This means in an emergency situation data is anywhere from 60-100% on the backup hard drives already.
Write Accelerator systems take reliability even farther by implementing IBM Chipkill technology. Standard error correcting code (ECC) implementations correct single-bit data errors in memory chips but cannot correct the multi-bit errors that can result in data integrity issues. Chipkill technology allows a memory system to correct a multi-bit failure up to and including a total chip failure.
All solid state disks can correct a single bit error before sending data to the server as a part of their ECC protection. The Write Accelerator, however, goes a step further by scrubbing, re-writing the corrected data to memory and then verifying the re-write to determine if a memory chip has a failure (a hard error) or if radiation transients caused the single bit error (a soft error). Systems that do not scrub single bit memory errors will either a) report errors to a system log that will eventually encourage replacement of a memory board or b) hide these errors thus leaving a potentially unsafe memory board in the system. Research on this topic suggests that 90% of single bit errors are soft errors. In these cases, the Write Accelerator will correct the error through the scrubbing process and prevent unnecessary downtime to replace the memory board.
