News

Microsoft Puts Reference Customer Forward for SQL Cluster

Microsoft Corp. has gone public with a production customer using its Distributed Partition Views technology in SQL Server 2000.

The reference customer, BrightStreet.com, comes after more than a year of criticism that the technology Redmond used to scale to the top industry-standard benchmark performance had no real-world adherents.

BrightStreet's relatively small 3 GB database demonstrates nothing about the scalability of SQL Server using DPVs. However it does address industry concerns about the complexity of managing a group of SQL Server machines as a single logical database.

"There was a learning curve there, but I tell you, once I got into it and found out exactly what was going on, I fell in love with it and ran," says C.J. Butcher, database software manager for BrightStreet.com.

Distributed Partition Views debuted with SQL Server 2000. The technology allows database architects to build a database across several server nodes. In the typical example, a large database would be split so that records A-F would be stored on the first node, G-L on the second node, and so on.

The approach allows Microsoft to overcome the limitation of its traditional cap of eight-processor servers by ganging several servers together for greater scalability. (Microsoft can now scale to 32 processors in a single server with Windows 2000 Datacenter Server, but the technology is not in widespread use and the systems cost more per processor than the standard high volume eight ways).

With the Windows 2000 launch in February 2000, Microsoft used DPVs to cluster 12 Compaq eight-processor servers and publish a record result on the Transaction Processing Performance Council's TPC-C benchmark for OLTP performance. Since then, Microsoft and its partners have improved upon the numbers, and Windows-based systems have maintained the top spot in raw TPC-C performance.

BrightStreet, which initially set up DPVs on two server nodes, provides what it calls a Digital Promotions Backbone. Customers, who include manufacturers, retailers and others with an Internet presence, can set up marketing campaigns through coupons, free samples and repeat-purchase loyalty programs targeted at consumers.

After the campaigns are set up, consumers responding to the promotions get routed through BrightStreet's infrastructure – a model with potential for heavy server traffic.

BrightStreet's Butcher, who has experience with SQL Server 7.0, Oracle 6, 7 and 8, Informix and Sybase databases, says he set up DPVs in about two weeks with no special help from Microsoft. "I was about ready to call them once, but then I found what I was looking for and didn't need to," he says.

Butcher had done some advance work, setting up Microsoft transactional and merge replication services across nodes in BrightStreet's back end that translated into configuring DPVs. "The replication was much more complicated. That was one of the reasons that we decided to go with the DPVs," Butcher says.

Jeff Ressler, a Microsoft product manager on the SQL Server team, says the BrightStreet experience shows competitors' criticism of DPVs is off base. "This is an example that Oracle's claims that this is not a real-world deployable technology are false," Ressler says.

Because the technology ships with every copy of SQL Server 2000, Microsoft has no way of tracking how many customers are using DPVs, Ressler says. Microsoft is aware of no more than 10 customers working with DPVs, he says.

BrightStreet's implementation shows that there are more reasons to use DPVs than immediate scalability, Ressler contends.

For BrightStreet, the approach guarantees room for future expansion, Butcher says. Each consumer that comes through BrightStreet's system gets a number. Originally, the company wanted to create an identity field in SQL Server 2000 for assigning those numbers, but the data field's 10-digit limitation was a problem.

"We wanted to dedicate a billion numbers to each server. That limited us to 10 servers," Butcher explains. "If you're looking at impression counts. We felt like that billion number might be approachable."

Going to the distributed views, it was possible to run up to something like seventeen quintillion unique numbers, according to Butcher. "We more or less felt like this option gives us scalability way out beyond what we'll ever achieve," he says.

Butcher does have a couple of suggestions for improvements to the DPVs technology. The indexing capabilities within DPVs could be improved and procedures for updating DPVs could be simplified.

BrightStreet had been a reference customer for SQL Server 2000 before. The company runs complex stored procedures every time a customer record is updated. Butcher reported a drop in execution times from four seconds to less than one second when the company upgraded its Windows 2000 Advanced Servers from SQL Server 7.0 to SQL Server 2000.

Butcher implemented DPVs in January and reports no downtime or extra daily administrative burden. He has plans to increase the number of nodes in use. -- Scott Bekker

Related Articles:
High-Availability Clustering Market Shrinks
Clustered Grids Make Their Way Into the Enterprise
ENT's Guide to Microsoft Clustering Technologies

About the Author

Scott Bekker is editor in chief of Redmond Channel Partner magazine.

comments powered by Disqus
Most   Popular