Oracle Database Management
Oracle gets chosen for applications that demand real-time responses on massive databases with thousands of users. Historically, the performance of these databases was so critical to business success that it was relatively price insensitive. Today, it can't be at any cost anymore — IT budgets are flat. Concurrently, the trend on transaction performance is down. What's happening?
- Big Data. The need to meet performance as databases increase in size is becoming a bigger challenge simply due to the shear size of the databases. As datastores increase in size, the supporting infrastructure needs to scale with it. While servers, and other semiconductor-based infrastructure have increased 175X in performance over the last 10 years, disk performance has remained the same.
- Customers are consolidating server hosts. As illustrated in the diagram below, when customers consolidate their databases onto fewer servers with larger CPUs, the resulting IO stream going to the storage array is far more random. This destroys the performance of a disk-based array because it has to seek for data on ever denser disks, with a fixed spindle-constrained bandwidth.
Our Oracle performance actually increased after moving to a virtualized deployment with Pure Storage
- Bryan Bond, Sr. Systems Admin, Siemens
Pure Storage handles the randomized IO that results from server consolidation easily because it combines the performance of flash with an unique manner of distributing, deduplicating, and retrieving data from the array. Running Oracle on Pure Storage allows you to:
- Accelerate Redo Log Writes (OLTP). Write intensive applications often experience log file sync wait events due to frequent transactional commits. Migrating high-commit rate application's redo logs to flash enables faster log buffer flush times which can dramatically improve application performance.
- Reduce Dependence on Index Scans (DW). Indexes are a key tool for improving sorted query performance, but sometimes require significant additional storage, as well as additional CPU and IO overhead to manage. Simply having an index sometimes influences the Oracle Optimizer to avoid direct path reads, thereby penalizing query results. Migrating your database to Pure Storage will significantly reduce the cost of disk based sorting in the absence of an index as compared to SGA latching overhead when traversing an index.
- Turbo-charge Temporary Tablespace. When additional sort/join space beyond the Process Global Area (PGA) memory allocation is required, Oracle looks to the Temporary Tablespace. Locating the Temporary Tablespace onto flash allows for a direct transactional performance improvement.
- Simplify Data Layout. As a DBA, you've probably been forced to get a PhD in storage, understanding in great detail how LUNs are striped across spindles in your storage architecture. Getting the best performance out of Oracle has required complex configuration of your underlying storage objects to create the right performance profile for logs, indexes, data files, and temp databases. You may have employed Oracle ASM for avoiding hotspots with marginal success. With Pure Storage, you can throw all that out the window. Create one big LUN, put everything in it. Done. Fast.
- Any Oracle Block Size. In addition to understanding your storage LUN layout, fine-tuning the Oracle block size and alignment has been critical to ensuring performance. Pure Storage's 512-byte virtualization architecture means block alignment issues disappear. Pick any block size, they're all fast.
- Virtualize Oracle Without Fear. Tier 1 databases typically drive highly random I/O, and are incredibly latency-sensitive...not the kind of application typically considered for virtualization, until now. The abundant low-latency random IOPS that Pure Storage delivers enables the virtualization of Oracle with ease.
- Reduce Data with Better Performance. Pure Storage's inline deduplication and compression seamlessly serve Oracle environments, allowing reduction of Oracle database physical storage requirements by 5x-10x with better performance than legacy disk arrays.