Wednesday, September 24, 2008
Oracle Exadata 'X' is out
Convergence of hardware and datab ase soft
Larry Ellison announed the database machine today in his keynote, a machine build on radicle new ideas. He used the sail boat analogy on the out of box thinking needed to overcome the bandwidth limitation between the storage system and the database. The large DW's are tripling in size every two years and are plagued with the data bandwidth problem. There is a disk to Database choking. Storage disks can easily store 100TB but movement of data is the biggest.
Most hard configurations slow down at even lesser values 1-10 TB range. The largest storage systems show the exponential increase in scan time at 10TB. This problem can be solved in two way, as a data problem, reduce the data - compression or indexing the data or partition pruning etc. The other way it to enhance the amount of data travelling over the data pipes. Make the pipes faster and increase their number.
This lead to the announcement of Oracle's first hardware product called Exadata programmable storage server in partnership with HP. This is a combination of hardware and software. The idea is to locate the intelligence next to the storage server, to process the query in the h/w side and ship only the results. Typically a query on a large table that is not indexed, results in full table scan, i.e. shipping of all the blocks to the database, this is the point of choking. To reduce that, now the intelligence in in the storage system that limits the results and only ships the result blocks to the database. This will help to improve the parallel query. Now the idea is resturn results not disk blocks. Query is now moved to a lower layer. There will be three grids now, grid of database, grid of fusion middleware and grid of storage servers. The connection between the storage server and DB is the Infiniband pipe at that can practically transfer data at 1 GB / sec.
The exadata is available for immediate shipping for Linux and will be extended for all OS. The Oracle database machine or 'X' was announced as the competitor for Teradata and Netezza platforms or the hardware database appliances. The results of three years of R&D leads to X as the fastest database machine on earth. It has 8, total of 64 Intel cores and 112 cores of Intel CPU just for storage. It can hold 168 TB of data and is 1000 X larger than largest Ipod.
The pilot customers were Yahoo, NPD (the company I cosulted at last summer when I was in Oracle's BI practice), Country wide, Retailer Giant eagle, Amazon etc.
The European Telecom provider, saw 10X to 72 x query performance 4.5TB of CDR's, average 28 X perf (on half the config of X). This is compared to 2 IBM p series and EMC disk array. The largest imporovement was in CRM report - customer discount report
LGR telecom DW for phone companies saw 30X performance improvement, compared to HP superdome + based config with Hitachi array.
Other Examples of Performance Improvement
Chicago Board of Trade 10-15 X perf
Giant Eagle Retail sales 16 X performance
Oracle's Internal financial data warehouse netapps 30X speedup
In a nutshell, X has intelligent storage and more bandwidth. Rather than conventional movement of disk blocks, X shows almost same access time as the DB's in size making is very scalable as DB and DW's grow into petabyte ranges.
Market comparision: X is faster than 5 rack Terdata 5550
Larry said that there is no query intelligence in Teradata, very sophisticated DB though, proven over many years. Teradata moves disk blocks.
Unlike Teradata, Netezza, is a storage server, however, in the comparision shows, it was about 2:1 improvement in data bandwidth (14 GB/sec compared to 7.5 GB/sec)
Besides, the Netezza h/w does not run Oracle database.
One one had the Teradata is a very good DB, netezza is not. X provides the balance between both the current market leaders for DW appliances.
Now let's look at the pricing model, $1,680,000 for the license for X, about 650 for the H/W, cost of about 4K/TB of data at undiscounted level. Its open technology, Intel CPU's with 6 cores can be used in future, will take advantage of
cpu and ram speed etc. reducing the cost to the end user.
Currently, 90 of the load in OLTP systems is queries and reports. Therefore, it is imporant to make sure that both databases for OLTP and data warehouses can be run on this platform.
HP and Oracle will together take orders for X, Mike Hurd of HP came "online"
Let's keep an eye how the rest of the world will respond to this news..