Oracle has set the bar extremely high for Exadata. We are designing a single hardware platform that is best for any class of Database Workloads. The hardware is identical for all workload types since high compute capacity, high network bandwidth, low latency, high IO throughput, and Flash help any workload. The software features vary by workload. For example Oracle has many Warehouse specific software features such as Bitmap indexing, integrated OLAP and Data Mining, etc. The Unique capabilities of Exadata are highlighted in the presentation by being underlined in Red
Oracle believes that the Scaleable Grid architecture is the architecture of the future. It eliminates the long-standing tradeoff between high-end hardware platforms that are scaleable (up to a point) and available, but also very high cost due to low volume. Sometimes people think that because there is so much compute and storage in a database machine it must consume huge amounts of power. In fact the power consumption is not large. Maximum Power usage of a Full Rack Database Machine is 14KW, typical is 9.8 KW. A single high end SMP platform without storage or switches can consume well over 20 KW.
Exadata allows easy expansion within and between racks. A quarter rack can be upgraded to a half rack. A half rack can be upgraded to a full rack. Two half racks can also be connected together to form the equivalent of a full rack. This is sometimes useful in data centers with weight or heat density restrictions. Once a full rack is deployed it can be increased in half rack size increments. For example a full rack grows to 1.5 racks, then to 2 racks, then to 2.5 racks. Equipment can be mixed across hardware generations. For example a half rack can be grown to a full rack using the next generation of servers to fill out the rack.
This is all about scale out. Scale outwards ! Exadata has much more compute, storage, and interconnect capacity than other systems on the market. An 8 rack Exadata system is equivalent to at least 30 racks of leading conventional products, and costs much less. To purchase equivalent compute and storage capacity using conventional equipment would cost over $100M (The price of a single IBM P595 taken from the TPC web site is about $4MM). The user data capacity is much larger after compression. Scales beyond 8 racks by using external InfiniBand switches
Eliminates the complexity of deploying a high performance database system. Database machines are tested in the factory and delivered ready to run. Because all database machines are the same, their characteristics and operations are well known and understood by Oracle field engineers and support. Each customer will not need to diagnose and resolve unique issues that only occur on their configuration. Performance tuning, and stress testing performed at Oracle is done on the exact same configuration that the customer has ensuring better performance and higher quality. Applications do not need to be certified against Exadata. Applications that are certified with Oracle Database 11.2 RAC will run against Exadata. Very few applications need to certify the storage subsystem underneath a database, and Exadata fundamentally is the Oracle Database with a very fast storage subsystem.
Exadata storage servers scan the data in parallel removing all central bottlenecks. Traditional storage has bottlenecks at the back-end disk loops, caches, controllers, front-end channels, and SAN links. In traditional storage, receiving Terabytes of data into the DB server and filtering it consumes large amount of CPU in the database hosts.
The Exadata Smart Flash cache avoids cache wipe-outs caused by large scans. A simple example is that Exadata knows when a read operation is caused by a backup, and avoids caching the blocks. Note that Enterprise Storage Arrays now support flash disks but there are no reported IOPs numbers from any vendor for their storage array using flash. The I/O performance numbers shown here are measured at the database level, not pure storage statistics that cannot be achieved in practice. Some vendors quote component level performance numbers that cannot be achieved in a complete systems due to bottlenecks at other parts of the system. Also, remember that Exadata is a full system including servers, storage, and networking, not a pure storage device when comparing to other products. NetApp and EMC have released flash cache products. EMC uses conventional flash disks for cache which makes them much slower than Exadata’s Flash PCI cards. NetApp uses Flash PCI cards for their cache, but the largest NetApp systems (6080) have at most 8 flash cards. Neither vendor is willing to quote specific IO’s per second or data throughput numbers for their flash solutions. Neither vendor can combine flash with 10x compression, InfiniBand, and storage offload. Only Oracle allows individual tables and partitions to be placed in flash with a simple command.
Exadata has fully redundant hardware. Redundant servers, redundant storage servers, redundant network. So any component can fail and the system as a whole will keep running. Our measurements to date show that the hardware failure rate is dominated by disk failures. The Oracle database software tolerates failures by continuing to run when various hardware components fail. For example Oracle RAC continues to run after server failures. ASM mirrors data across storage servers so that the failure of a storage server does not cause an outage of the system as a whole. Oracle has unique capabilities for rolling back erroneous changes called flashback. Oracle has unique capabilities for making changes to databases online called online redefinition. All truly highly available systems should have a remote replica. Oracle has the industry’s leading technologies for creating and maintaining remote replica databases. Golden gate provides a powerful symmetric replication capability. Active Data Guard provides an extremely high performance and simple way to create a readable remote replica database.