http://www.itjungle.com/breaking/bn071107-story01.html

Oracle Says 11g Database Is Better, Cheaper, and Faster

Published: July 11, 2007

by Timothy Prickett Morgan

Charles Phillips, president of database and application software supplier Oracle, made a prediction at the launch of the company's 11g relational database in New York City today. "What normally happens in this process is that, initially, there will be many articles written about how no one needs another version of the database. The pundits are already writing that, and it is typical." But software companies add features to software and change it to address new demands, and Oracle is no different.

With the Oracle 11g database, over 1,800 developers inside the company have created over 400 features, driven by the needs of customers, and tested them extensively to create this new version of the company's relational database--which is in use at 275,000 companies worldwide and which gives Oracle the dominant position in the relational database market worldwide in terms of revenue share according to statistics from Gartner. As was the case in past releases of the Oracle database since Oracle 2 was launched 25 years ago, Oracle is fairly confident that many of the new features will often individually drive existing Oracle customers to adopt the new version, as well as making new customers to give it serious consideration over alternatives from IBM, Microsoft, MySQL, Sybase, Ingres, and a handful of other companies that do commercial, general purpose relational databases.

Like the operating systems they run on, databases change very slowly, and necessarily so because of the myriad ways that they are used by customers and the inherent complexity of the software itself. It takes a database maker like Oracle years to assess the requirements that its customers have and to analyze what the competition is up to and then to weave new features into the code. The last big upgrade to the Oracle database family was Oracle 10g, which was launched in October 2003 and which was updated with 10g R2 in the summer of 2005. And to be fair, Oracle 10g was just an extended version of Oracle 9i RAC--short for Real Application Clusters--packaged up differently and extended. Oracle 8i, launched in November 1998, was the first real big change in the database in as much as Oracle had firmly embraced Java and had conned Compaq into giving it the parallel file system guts inside of its own Tru64 Unix to create the RAC extensions that were eventually added to Oracle 8i and polished with Oracle 9i.

With Oracle 11g, the concept of grid database computing--using parallel computing nodes to run separate yet connected instances of databases with shared storage--is being tweaked a bit. But more importantly, Oracle has done a lot of work to make the database easier to manage and cheaper to deploy--both of which end up being money saved in the long run.

One new feature that allows standby servers running replicated databases (using Oracle's Data Guard feature) to be put to work doing backups, reports, and such, freeing up capacity on production machines (so they can do more work). Oracle has also created features that make it easier to create test environments and run real production transactions against new Oracle database releases, thereby getting customers to adopt the software more quickly in their production environments. This latter feature, called Real Application Testing (no, Oracle does not want it to be abbreviated Oracle RAT), can reduce the typical test cycle for database applications from around 150 days to around 11 days. This is a huge savings, especially when you consider that the real reason why EMC's server virtualization subsidiary, VMware, is a contender in the data center is that its Workstation, ESX Server, and Lab Manager products shorten the development and test cycle. People talk about server consolidation a lot, but dev and test is what has been driving a lot of virtualization installations on X64 iron. (That's changing, of course.)

Another set of money-saving features relates to the partitioning capabilities inside the database itself. For years, Oracle's databases have been able to be carved up into multiple but related tables that, through the magic of software, still look like a single table to the database management system and the applications that get data through it. Data can be partitioned according to any number of indexes, which allows queries on subsets of data--sales data in a specific geography--to run a lot faster than it would against a larger table. (Soap and toothpaste giant Colgate Palmolive has over 1 million partitions in its databases, so it can do fast queries against the data relating to its myriad products, and many smaller customers have tens or hundreds of thousands of partitions.) But creating and managing these database partitions is a totally manual process. With Oracle 11g, a partition advisor watches how the database is used and recommends specific partitioning of the database and the database can create partitions as the indexes for the partitions change--for instance, as each month opens for data that is partitioned monthly.

Another useful feature in Oracle 11g is integrated compression, which can compact files by a factor of two or three, and sometimes as much as five in Oracle's initial tests with customers. According to an analysis presented by Andy Mendelsohn, senior vice president of server technologies at Oracle and the person who has been involved in database development since Oracle 5.1 back in the mid-1980s, the amount of data that companies are storing inside their databases is tripling every two years. And if there is one thing they want, it is to spend a lot less money on disk storage for databases. So with the integrated compression feature, coupled with features in the database that automatically move data from high-performance to midrange to slower disk arrays as it gets used less frequently, Oracle can substantially reduce the cost of disk storage.

In the example Mendelsohn showed, a company wants to deploy 14 TB of database files and needs $1 million of high-performance disk arrays to do the job; this works out to about $72 per GB. However, knowing how applications actually use data, Oracle says that only 5 percent of the capacity of the database needs to be on the fastest disk arrays, while about 35 percent of the data, which is accessed occasionally, can be moved to midrange disk arrays that cost maybe $14 per GB. That leaves the remaining 60 percent of the data that can be moved to slow, cheap disks. When you do this kind of partitioning, you can reduce the cost of disk arrays to under $200,000, and when you do compression on the files, you can get it down to $60,000. That is a lot better than $1 million.

Oracle 11g will also support new data types, including native RFID data, medical records stored in the DICOM industry format, and 3D spatial information. Oracle has also created an improved XML format called binary XML, which complements the XML large object binary (LOB) and XML object relational formats already supported in Oracle databases. The upshot of binary XML is that it can cope with LOBs about 15 times faster--which is important for new kinds of media data. The upcoming database also supports something called fast files, which can deliver database reads that are as fast as the Linux file system itself and database writes that are actually faster than Linux. There is also a neat feature called total recall that allows a query to be executed in a database today as if the database were from a point in the past. So you can ask questions about the past in your database without having to do complex analysis.

On the performance front, 11g has been put through some tests and query performance on Oracle RAC grids has been improved by 70 percent over 10g R2, and the Java just in time compiler has had its performance boosted by a factor of 11. Oracle streams replication (which is how data is moved around the grid) is twice as fast, and query caching is up to 25 percent faster.

Oracle 11g will be available in the Express edition, which is free to developers, as well as Standard Edition One for entry servers, Standard Edition for midrange boxes, and Enterprise Edition for bigger boxes and machines that need all of the most sophisticated features of the database. Oracle sells its database software by processor core (using an algorithm) and by end user.

Oracle's database packaging and pricing is not expected to change with 11g, at least not for the core database products, according to Chuck Rozwat, executive vice president of server technologies at Oracle. Neither he nor Phillips would say when Oracle 11g will ship, except to say that the version of the product for Linux will come out in August and that detailed pricing for the different 11g editions would be finalized as that launch date approaches. No Oracle executive would say when Unix, Windows, or mainframe versions of 11g would be available. Rozwat also left open the possibility that Oracle might charge separately for some of the new features. Customers on maintenance contracts with prior Oracle databases will be able to upgrade to the new program as part of their contracts.

The question, of course, is will customers go for the new database, or stick with what they have? According to Ari Kaplan, president of the Independent Oracle User Group, members surveyed about their 11g upgrade plans are pretty keen on the technology. According to its survey, 35 percent of customers polled said they would move to 11g in the next 12 months, while over half said they would make the move in the next two years. The uptake for Oracle 10g happened faster than expected, and given the feature set in Oracle 11g, the company is expecting an even faster uptake for the new product, according to Phillips.

Back to Ari Kaplan's Home Page