Last week got an opportunity to attend a workshop on Oracle Exadata. About two years ago, when Exadata first released, attended another introductory session presented by Oracle Asia experts. That time I was a bit hopeless with the product as Exadata was specially focused on very big Data Warehouse solutions. I was expecting something that fits with our OLTP + mixed kind of systems and may benefit us.
Now, with the Exadata version 2, it seems that they are probably going to impress everyone in terms of performance and scalability. I found it very exciting this time!
The Exadata version 1 was marketed for Data Warehouse type solutions with very big system - 8 nodes Database Server Grid and 14 nodes Storage Grid (which they are calling now Full Rack). For a small to medium kind of system, this was not perfect and efficient (in terms of money) I believe. After acquiring Sun, Oracle came up with Exadata version 2 with Sun hardware (version 1 was on HP) and emphasis given on OLTP systems as well. More importantly, they came up with different packaging solutions like Half Rack and Quarter Rack which are a kind of thing I was interested in.
The cool features of Exadata are -
* Exadata Query Offloading: Exadata solution offloads data-intensive query processing from Oracle Database servers and does the query processing closer to the data storage. The result is faster parallel data processing and less data movement through higher bandwidth connections. This massively parallel architecture also offers linear scalability.
* Smart Flash Cache: A transparent extension of database buffer cache. Lot more I/O throughput and huge performance improvement. 1 GB costs only $50. We would be able to cache the whole database almost. They say, "This will break random I/O bottleneck by increasing I/O per sec by 20X and doubles user data scan bandwidths".
* Hybrid Columnar Compression (HCC): They say, "Efficient compression increases effective storage capacity and increases user data scan bandwidths by a factor of 10X". Online archiving compression feature will reduce our historical data size about 90%. This will help us to manage old data in one place. No need for separate historical data servers in cheaper disks. Old data partitions will be compressed in online archive mode.
* Exadata Storage Indexes: Eliminate unnecessary I/Os to disk and contributes to the overall performance improvements. These are not actually index, these are kind of filters which intelligently organizes data inside disks. Result is big performance improvement - even with full table scans!
I made a presentation for my other team members to give an overview on Exadata. Here is the summary -
Oracle defines Exadata like this -
* Massively Parallel Grid, the Architecture of the Future
* Extreme Performance: 20X for OLTP and 10X - 100X for data warehouse
* Linear Data Scalability: Same performance for 1 TB and 5 TB of data
* Enterprise Ready: Packaged, readily available. Single support from Oracle for hardware and software
* All Standard: Works transparently with all existing applications
Improvements:
* Millions of transactions per minute
* 10 millions queries per minute
* Billions of rows per minute
* 1 million I/O per second
* 50 GB/sec flash bandwidth
* 21 GB/sec disk bandwidth
* InfiniBand network: 40 Gb per second throughput
Different Exadata Appliances
* Full Rack: 8 Database server and 14 Storage cell server
* Half Rack: 4 Database server and 7 Storage server
* Quarter Rack: 2 Database server and 3 Storage server
* Single Server: 1 Database server and 1 Storage server
Exadata Quarter Rack
* 2 Sun Fire™ X4170 Oracle Database servers
* 3 Exadata Storage Servers (all SAS or all SATA)
* User Data: 6 TB SAS or 21 TB SATA
* 1.1 TB Flash Cache (4 x 96GB PCI Express Flash Cards on each)
* 2 Sun Datacenter InfiniBand Switch, 36-port Managed QDR (40Gb/s) InfiniBand switch
* 1 “Admin” Cisco Ethernet switch
* Keyboard, Video, Mouse (KVM) hardware
* Redundant PDUs
* Single Point of Support from Oracle
* 3 year, 24 x 7, 4 Hr On-site response
Performance Benchmarks
Some of the performance benchmarks shown -
* 1 TB tablespace creation time was about 9 min! On other hardware platforms it will take about 2 hours!
* On full rack, the test query with ‘query offloading’ turned “off” (means simple RAC), took more than 4 min to execute where turning query offloading “on”, took about 5 sec for the same query!
* Another Data loading example shown - 49 GB data loaded from external file in just 1.5 min!
***
