Almost one year after Oracle and Sun introduced Exadata Version 2, user interest is starting to turn into purchases.
During its quarterly earnings call, Oracle said its Exadata Version 2 "pipeline" -- which refers to estimates of future sales -- was approaching $1 billion for 2011, and the database and online transaction processing (OLTP) machine already includes customers such as Bank of America, Carrefour and Thomson Reuters.
That stands in stark contrast to the first version of Oracle Exadata, an HP machine that was estimated to have sold only a couple dozen units. Oracle is pushing Version 2's OLTP capabilities in addition to its data warehousing feature, and the company has said the engineering work it did with Sun has paid off. That joint work didn't use either Sun's SPARC processor or its Solaris operating system, but rather Intel chips and Oracle's Enterprise Linux operating system. Consultants and end users said that doesn't even matter.
”My opinion is that the commodity architecture is what makes Exadata affordable in terms of hardware pricing," said Alex Gorbachev, chief technology officer at database consultancy Pythian. "But technically there is nothing that could prevent Oracle from putting SPARC inside.”
Oracle Exadata Version 2 pricing isn't exactly cheap, though, and it can easily reach more than $1 million before support and licensing costs kick in. When all is said and done, companies that install fully loaded Exadata machines could easily end up paying several million dollars.
Others have said the key to Oracle Exadata performance is not in the CPU but in the machine's software, which allows a lot of the work to be done at or close to the storage level.
LinkShare, a New York-based Internet marketing company, has installed two Exadata machines, one a full rack and the other a half-rack. They did it with the help of Pythian's database consulting services. The company works with clients to place advertisements online and then analyze how well the ads performed. That analytical system, which reports results and helps LinkShare and its clients decide what to do next, will be the main program on its Exadata machines.
LinkShare received its Exadata machines in March and plans to have them running in live production next month.
”Our old platform is a very large Linux-based shared-nothing database," said Jonathan Levine, LinkShare's chief operating officer. He added that the system ran on commodity hardware running Intel chips. Levine declined to say what the data warehouse and database was, but a case study from 2007 says it was running IBM's DB2. "We found shared-nothing architecture very difficult to scale across large numbers of partitions," he said.
LinkShare was running about 18 physical nodes and almost 70 vertical partitions, according to Levine. The communications among the partitions was via Ethernet, and Levine said they were often "saturating the backplane" when doing work across multiple nodes and partitions. Oracle Exadata Version 2 distributes querying across compute and storage nodes, easing that problem, he said.
"Joining multiple big tables was a big issue, " he said. "Large amounts of data have to move from the storage nodes to the compute node, and that becomes a network issue. Exadata does a very good job pushing down parts of the query to the storage cells, limiting the amount of data that has to be sent to and processed by the compute nodes. Together with the faster interconnect, this lets us analyze more data faster with less power and in less space."
Kerry Osborne agrees. Osborne is the co-founder of Enkitec, a Dallas-based Oracle consulting firm that has recently installed an Oracle Exadata machine for testing and proofs of concept for clients. Osborne has been blogging about the adventure and including details about the company's implementation.
”Basically, it's spreading out the work onto a bunch of mini-SANs that can do the work at the storage level and then send the results back to the database server," he said. "It's sort of parallelizing all the work. So I don't think putting faster CPUs in this situation would make that much of a difference."
Oracle Exadata Version 2 also has the added benefit of running OLTP transactions, while the first version was only for data warehousing. Gorbachev, the CTO of Pythian, said that is largely a product of Oracle's embrace of flash memory cache technology, which essentially replaces slower I/O operations to disk with faster flash memory operations. That, according to Gorbachev, makes for faster transaction processing.
"The flash cache is one of the more important aspects," he said.
Because the first version of Exadata was geared just for data warehousing, interest "just wasn't very widespread because the market was fairly narrow," according to Osborne. That has changed with Oracle Exadata version 2.
”Almost everyone we talk to is interested in it," Osborne said. "Of all the new releases Oracle has done in the last 20-plus years I've been working with them, this has brought more interest than anything I can remember."