I've been encouraging Oracle to serve up Exadata reference customers instead of sales pipeline claims. The company delivered early this week, setting up an interview with Paris-based financial services giant BNP Paribas.
In my interview on Tuesday, Jim Duffy, BNP Paribas' data warehouse architect, went into great detail about the deployment.
Here are a few highlights:
- BNP Paribas is using a half-rack deployment of Exadata V2 to replace a four-node RAC cluster running Oracle 10g. The deployment went live two weeks ago, and like the system it replaces, it serves as the underpinning of some 35 trading-floor applications. That includes near-real-time performance stat monitoring, risk- and market-abuse compliance reporting, network and internal application optimization, and long-term (five-year) archival compliance querying and reporting. A second Exadata V2 half rack is dedicated to disaster recovery.
- Exadata V2's Smart Flash Cache -- 5 terabytes' worth in BNP Paribas' case -- supports the high end of the company's high/medium/low-speed data-access scheme. The last week's worth of trading data is stored in flash cache, as is anything directly accessed by an internal-facing Web-site used for monitoring and querying (particularly derived data).
Cache is also used for staging tables that are accessed directly to deliver internal application performance statistics. Duffy said the monitoring/querying Web site is serving up answers five times faster than it did with the RAC/10g deployment.
- The total warehouse now stores less than 10 terabytes, down from 23 terabytes in the previous environment. Duffy credits Exadata V2's hybrid columnar compression with cutting the database down to size (and reducing storage and admin demands in the process). "I can get tables that were already compressed down to 1 terabyte in Oracle 10g down to 185 gigabytes using the hybrid columnar compression in Exadata," Duffy said. "That brings massive benefits in terms of manageability."
- Elaborating on admin and management, Duffy said he had to tune the old RAC implementation "six ways to Sunday" to maintain performance while coping with ever-changing database loads. With Exadata V2, the peaks are easy to handle. Maintenance tasks that used to take hours can be done in minutes on Exadata V2, and data modeling steps that used to take more than two hours now take less than 20 minutes, Duffy said. "More than half the time we spent tuning vanishes, and all that time can be immediately dedicated to development," he said.
- The deployment does not address the investment research side of the business -- the part that analyzes market opportunities, trades and risks (other than those related to compliance).
- Existing queries are running about 16 to 17 times faster on Exadata V2 than in the old RAC deployment, on average. Duffy said he hasn't had time to change any code to further exploit Exadata V2's ability to push SQL querying down into the storage tier, which could improve performance on queries accessing lots of data.