The case for commodity hardware vs. Oracle Exadata architecture
A former Oracle sales rep does a 180 by pushing his company not to buy an Exadata system, but to upgrade existing data center infrastructure instead.
LAS VEGAS -- Cliff Burgess spent 12 years in Oracle pre-sales, trying to sell customers on the company's products and licensing.
But when he became IT director at Gentex Corp., an IT shop flush with Oracle Corp.'s products, two years ago, he did a 180-degree turn.
There is a false perception with Exadata that 'you install it; it manages itself; it bakes you cakes and everything else.'
The company was traveling swiftly down the path of purchasing Oracle's engineered system, Exadata, a combined hardware and software appliance. Despite Burgess' many years selling the Oracle vision, he told Gentex executives it made little sense to buy the costly Exadata system. Instead, he argued for upgrading the existing data center infrastructure with additional commodity hardware.
The decision not to go with Exadata likely saved the company $2 million in new hardware and software licensing costs, he said.
"We did the cost comparison of what we ended up investing versus what we would have invested, and it saved us $2 million," Burgess said during his session on Exadata alternatives at Collaborate 14, an Oracle user group forum.
Burgess' company, which manufactures automobile mirrors, as well as auto-dimming windows for aircrafts, processes 1.5 million transactions a week, so its Oracle database is transaction-heavy and subject to serious performance problems.
Oracle Exadata is a composite database server machine that uses Oracle database software and Sun Microsystems (acquired by Oracle in 2010) hardware. Oracle has touted the in-memory database appliance as a way to improve performance for transaction-heavy database systems, and to improve manageability by giving customers "one throat to choke" in the event of issues.
But IT director Burgess was more skeptical about the need for a high-priced system to solve the company's wait times and myriad performance issues.
"They were convinced that [Exadata] would be the silver bullet," he said. "You install it; it manages itself; it bakes you cakes and everything else."
Engineered systems proof of concept: The anti-Exadata business case
Building the business case for upgrading commodity hardware -- Gentex uses Cisco and Hewlett-Packard blade servers -- took time and effort from staff and third parties, Burgess said.
"We asked our partners to do a proof of concept with Exadata," Burgess recalled. "It was a very intense process. We said, 'We'll give you our data, go run it. Give us run times, and see what happens. We asked our other software partners [Cisco, NetApp] to load hardware and they did a similar venture."
Gentex ultimately opted to boost RAM and processors on its existing data center infrastructure, going from 36 GB of RAM to 256, but only a 50% increase in CPUs; Gentex went from 16 to 32 processors, which Burgess said saved the company $2 million in expenses for Exadata.
Gentex also split the storage tiers with production in one tier and non-production in another. The company virtualized its applications, as well. But given that Oracle doesn't support virtualization vendors such as VMware, licensing snags could pop up.
"You have to be careful how you architect your infrastructure so [you] don't expose yourself to licensing problems," he said. Gentex minimized its licensing costs by building clusters and buying some additional hardware to accommodate those clusters. It has virtualized nearly all applications other than its production database tier.
For Gentex, there were pros and cons on each side that needed to be weighed. The upside is that combined hardware and software of engineered systems creates a one-stop-shop to improve manageability and to "eliminate the finger pointing" where databases and hardware are siloed, Burgess said. The downsides include lack of knowledge among his staff on how to work with the system, high up-front costs, and the ability to address only some issues associated with database performance -- issues that might be attributable to other factors, such as bad code or the need to purge files.
However, commodity hardware is not without problems, either. Continuing to work with existing data center infrastructure means working with multiple vendors rather than benefiting from the one-throat-to-choke strategy of engineered systems, Burgess said. The commodity-hardware path also requires spending additional money on hardware to get solid performance and to manage the difficulties of navigating the complexity of Oracle licensing.
After some extensive proof of concepts by Gentex's hardware and software partners, Burgess and his 40-member IT team "were able to replicate Exadata performance with commodity hardware," he said. "So the question became, 'Why spend the extra cost, effort to bring in a new platform?'"
"We don't need to throw a lot of heavy iron at this thing to make things work," he said.
Is the path to Oracle Exadata inevitable?
Burgess' decision to boost performance with existing data center infrastructure has allowed him and his staff to establish a baseline -- "the new normal" -- and work on tweaking performance through other techniques, such as SQL scripting and file purging. Cleaning up data center infrastructure has allowed database administrators to focus on trouble spots and on boosting performance for the long term.
Oracle Collaborate 2014 conference coverage
Different users need different BI tools
NOSQL vs. RDBMSes: Handling the data onslaught
"We actually found a process that [has] been running for years … that we think is probably responsible for 50% of our IOPS. We never found it until we got cleaned up in the infrastructure. It was just lost in the noise," he said.
Still, while the revised Gentex strategy has staved off an Exadata purchase for the time being, Exadata may make sense in the future, Burgess said.
"We may go to Exadata at some point because … we need shorter runtimes," he said. "Just exhaust all your other options before you throw more hardware at [the problem]. … That hardware comes at a cost in terms of support, maintenance, more licensing and more complexity. We didn't want it to be the first thing we tried. We wanted it to be the last thing."
For more on the Oracle Collaborate 2014 conference, check out our conference coverage.