“If you come to a fork in the road,” baseball great Yogi Berra famously said, “take it.” That quote is metaphorically apt in contemplating the choice faced by the CME Group, which owns and operates the Chicago Mercantile Exchange. Will they continue their IT journey with traditional, rack-based servers, or accelerate their enterprise by embracing the bold path of engineered systems?
On a daily basis, CME Group executes, clears and settles millions of futures and options transactions, and manages hundreds of terabytes of data storage. In fact, recently CME Group had a record trading volume day, which consisted of nearly 27 million contracts traded. “It’s not a question of how many [transactions we can clear], it’s about how quickly we process them ,” said Walt Litzenberger, executive director of database systems at CME Group. “It’s really simple. It is critical that we are able to do our processing overnight, otherwise we will impact our customers SLA’s.”
That’s a tall task, because, as Yogi might have put, CME Group is “the biggest financial exchange you’ve never heard of.” (Actually, The Economist said that.) CME Group owns the Chicago Mercantile Exchange (CME), the Chicago Board of Trade (CBOT), the New York Mercantile Exchange (NYMEX), the Commodity Exchange (COMEX) and the Kansas City Board of Trade (KCBOT). In addition, CME Europe will launch later this year in London.
CME has been publicly honored for the past nine years with a place on the InformationWeek 500 list of top IT shops. And ComputerWorld recently named the company one of the 100 best places to work in IT for the ninth year running.
The IT capabilities of the group shine in its CME Globex electronic trading platform, which is among the fastest global electronic trading systems for futures and options. To accomplish the heavy duty online transaction processing (OLTP) requirements, CME Group decided to replace most of its rack clusters with 13
At this point in today’s post, you’re probably wondering why you should care about a case study of one particular enterprise’s success. It’s because CME Group’s experiences spotlight and reinforce the inescapable fact that engineered systems are the wave of the future. That fact that they’re here today constitutes an inflection point for CIOs, who must consider the strong argument that they help maximize the value IT delivers to your business.
That’s an assessment CIOs can’t avoid. Back in the day, the concept of combining hardware and software—i.e., considering how to optimize each to work better with the other, with the net result being faster overall performance—was a non-starter. Compute cycles (in plainer English, processing power) went only so far. Often software had to pull back, so that it didn’t overstress systems. As well, engineering culture kept software and hardware engineers on separate sides of a fence. (Because each considered their discipline the most important , that fence made for good neighbors.)
But the maturation of the microprocessor changed everything. As I noted recently in “Top 5 Processor Myths”:
“The years from the first microprocessor in 1974 up until the beginning of the multicore revolution in 2004 were consumed by the iterative maturation of a young technology. You gotta walk before you can run. When processor design and manufacturing finally advanced to the point where everyone knew how to make chips fast, other paths of innovation opened up for exploration.”
These paths have encompassed a plethora of performance-advancing efforts—and, again, here’s the important point—separate and apart from increasing the clock speed of the processor. Included are concerted efforts to drive:
- Deep multicore: For example, you can get an Exadata Database Machine X3 equipped with two servers, each of which has eight 10-core
Intel Xeon E7 processors. That’s 80 cores per server, or 160 cores total. - Rapid Data Retrieval: When processors were mad slow at executing instructions, there was ample time to get the next line of code queued up, and to read and write data from memory. Today, getting at data is the biggest deal. That’s why the Exalytics In-Memory Machine can grab terabytes of data off of disk storage before it’s needed and put it into flash memory, where it’ll be available for fast access.
- Power and Cooling: Electricity bills may be boring, but saving money isn’t. That’s why, for the past decade, reducing the number of watts consumed by your servers, as well as the air-conditioning cost of keeping them cool, has been high on every IT manager’s list. Consolidating, or cutting, the number of racks in your data center continues to be the most effective means of accomplishing this.
- Purpose-built, compatible storage: This is another Rapid Data retrieval value proposition. The Exadata ecosystem offers a separate storage building block—called, appropriately enough, the Exadata Storage Server [PDF download, here]. It’s tuned for use with the database machine, incorporating the same smart flash-memory cache, and exploiting massive parallelism, to cut read and write times to the bone.
Indeed, when you’re talking multimillions of transactions per day, accessing and archiving all that data is effectively equal in importance to processing power. “Exadata is the new Tier 1 in storage,” Troy Reece, senior director of technology at CME Group, told me. He added that it has replaced about half of the CME Group’s storage area network (SAN).
As our interesting conversation wound down, I asked Litzenberger and Reece whether, during the buy cycle, they had assessed competitive systems. One of these critical requirements was the ability to accelerate performance via parallel processing, or concurrency. It enables multiple transactions to be processed simultaneously. “When you get to concurrency, Exadata was the preferred provider to meet our needs,” Litzenberger said. “With Exadata, we can run up to 90 concurrent processes.”
On the question of perceived vendor lock-in—a perennial concern in the enterprise—I asked if that was something they had thought about. “Of course it was,” Reece said. “But do you need the speed, or don’t you? What does it cost you not to do it? Our expectation is that with each release of Exadata, it becomes faster and faster.”
What’s your toughest OLTP challenge? E-mail me here or follow me on Twitter at @awolfe58.
Further Reading from OracleVoice:
Complexity Barrier Makes IT Matter More Than Ever
Customer Experience Key To 'Computing Everywhere' Business Success
