At the recent Oracle OpenWorld Conference, DIY commoditized, best-of-breed data center infrastructure was out....
By submitting your email address, you agree to receive emails regarding relevant topic offers from TechTarget and its partners. You can withdraw your consent at any time. Contact TechTarget at 275 Grove Street, Newton, MA.
Pre-integrated, converged, virtualized, self-service provisioning, private cloud-in-a-box systems were in.
Users are weighing whether the tradeoff for ultra-engineered, “self-managing,” integrated, souped-up machines is worth the potential for lock-in, and the challenge of managing what’s essentially black-box technology.
Some IT managers have dubbed these boxes the new mainframes: servers, storage, networking, operating systems, hypervisor, management tools and middleware all pre-integrated, and each layer of the stack is optimized with the vendor’s secret sauce.
Sound like a mainframe yet?
“But it’s worse than that, it’s a return to the VAX,” quipped John McCarthy, VP and principal analyst at Forrester Research on the Oracle Exadata machine. “It’s the hardware, the OS, the database all combined. They’re putting all the wood behind one arrow.”
C-level execs may love them, IT admins fear them because of the potential to eliminate jobs and vendors are foaming at the mouth to sell them because they come with a significant cost premium over al a carte, commodity components.
The upside of converged systems: Performance and reduced head count
While some IT pros fear lock-in, others see the upside. Doug Miller, director of global database development at auto information broker R.L. Polk, runs an Oracle Exadata, and he said he could never build the kind of performance he gets out of the Exadata machine out of commodity parts.
“You can’t build this yourself. The secret sauce is the Exadata software, how it offloads the SQL operation to the storage cell,” Miller said. “Any large Oracle shop is going to get one of these, or its grandson. These will be like the IBM mainframe back in the 1970s. This is what you have and everyone has one.”
These converged systems can offer huge performance gains and reduce management workload because a single vendor owns and optimizes all of the components.
R.L. Polk, for example, provides customers access to large data warehouses. The data is published a couple times a month, and right after publication, a couple thousand users log on to get the latest figures.
“It’s not that busy all the time, but you have this peak time when it gets slammed,” Miller said. “Our customers’ expectations on how fast these reports should run keep going up, based on their experience with the Internet. If a user requests any complexity on the reporting, the response time is into the minutes. We couldn’t get the performance.”
RL Polk bought an Exadata V2 machine in July, a half rack with 18 terabytes of usable space. Miller said some queries have gone from two to three minutes down to five to 10 seconds. “Not every query runs 10 times faster than it did before, but 80% of them do.”
The other advantage for some companies is potential reduction in head count, or at least in inter-departmental wrangling.
“Building systems in the past was always a political problem,” Miller said. “There were so many headaches with different people in charge of the storage, people in charge of the SAN, sys admins building boxes. It was always a battle; I was never happy with the storage they set up. They were always complaining my requirements are unrealistic. I think about how much time we spent shopping around for what disk, what SAN switches, what server, what HBAs -- that’s gone. That’s part of the appliance argument. They built it and you don’t have to think about it.”
Miller said his network guys were initially bent out of shape when they found out the Exadata uses Inifiniband interconnects. “But they don’t even have to touch the Infiniband,” Miller said. “Do you log in and patch your Tevo box? No. Everything comes down from Tevo; you don’t touch a thing. It’s a box. Do you get upset because Tevo doesn’t run on Solaris? No.”
The downside of converged infrastructure: Lock-in
For companies that are largely single-vendor IT shops, these types of systems can make a lot of sense. But the majority of data centers remain a hodgepodge of data center technologies.
“When I talk to enterprise computing customers, I don’t see a lot of folks clapping their hands and giggling at the thought of having one vendor to work with,” said Charles King, principal analyst with Pund-IT Research. “Those companies have tended to buy whatever happens to be the best deal at the time. Single-vendor enterprise data centers are the exception, not the rule, and there may be some inherent problems working with multiple vendors, but that’s the model customers have chosen over and over again.”
Case in point, R.L. Polk is almost totally an Oracle shop. “Some people hate that. Some people hated IBM when they were the only game in town, but there are advantages,” Miller said. “We know Oracle is going to optimize middleware on this machine, and for competing third-party products it’s going to get tougher and tougher.”
One notable exception to this single-vendor trend is the Vblock joint venture by Cisco, EMC and VMware. These vendors pool support under a single organization called Acadia, or the Virtual Computing Environment Coalition. The VCE gives users a single throat to choke on support, and three companies are testing product performance and integration because all of their names are on it.
Dell is also skeptical of the bundled packages being marketed by its competitors. Dell’s data center consulting services are building and deploying custom bundled products for large customers, but a Dell rep at Oracle OpenWorld said that “off-the-shelf bundles look pretty but don’t get deployed very often. Everybody wants to customize.”
Another show attendee, a Boston-area IT tech consultant, said he’s seen quite a bit of interest in Oracle Exadata but is unsure how much of it will translate into actual sales.
“I have a client who‘s running an Exadata proof-of-concept and he says it screams. But he also says his company doesn’t want to spend the money and would rather spend a percentage of that cost and settle for less performance.”
Barbara Darrow contributed to this article.
Matt Stansberry is Executive Editor of SearchDataCenter.com. Let us know what you think about the story; email Matt Stansberry.