MarketWatch.com Inc. promises to provide accurate, up-to-the-minute financial information to businesses around
the globe and to do it 24x7. That promise was being broken, and the dot-com's storage system was to blame.
For MarketWatch.com, always-available database storage is critical for two reasons. First, customers in many time zones access stored information on its three Web sites: cbs.marketwatch.com, bigcharts.com and U.K.-based ftmarketwatch.com. Secondly, a global team feeds up-to-the-minute economic financial data, news and research information to the Web site.
"With 100 journalists in 11 bureaus around the world, as well as financial data coming in from around the globe, we can't afford any downtime," says Patrick Parker, MarketWatch.com director of data systems.
Downtime plagued Parker in late 2000 and early 2001, however. Back then, MarketWatch.com's RAID storage system crashed as often as every two days. "I was losing sleep worrying about crashes," Parker admits.
Even worse, the crashes were never easily remedied. "What should have been a simple crash and reboot usually turned into a daylong scramble to get things up and running again," says Parker. Any data stored in memory and not written to disk at the time of a crash would be lost, and the database would be corrupted.
"What I discovered was that anytime that SCSI card failed, I couldn't get to my data," he recalls.
MarketWatch.com had purchased this private-labeled RAID system from one of its systems vendors, whose specialty is not storage. The vendor's in-house RAID expertise was limited. "Our vendor had no answers and wasn't willing to step up and fix the problem," Parker says.
Faced with an untenable situation, Parker looked for alternatives. At first, he considered clustering database servers. "I decided it could set me up for a bigger catastrophe," says Parker. "With clustering, we would need to set up two storage systems for every cluster. So, if one went down, we'd have another one to go up."
Having researched Fibre Channel-based storage alternatives a couple of years ago, Parker was impressed with Fibre Channel SAN's speed and high availability. He'd wanted to choose it before, but the company's database - Microsoft SQL Server - didn't fully support Fibre Channel back then. This time around, in 2001, Fibre Channel and Microsoft SQL Server were compatible.
After evaluating several Fibre Channel SAN products, Parker was won over by the reliability, performance, ease of expansion and price of Carlsbad, Calif.-based Dot Hill's SANnet storage systems. A Fibre Channel RAID Storage system, SANnet promises 99.999% reliability. The SANnet 4000 Series Fibre Channel RAID devices offered redundant Fibre Channel connections, which support one or more servers simultaneously.
The capabilities of Dot Hill's SANscape software fit MarketWatch.com's need for simple administration and management. For example, in event of a SCSI card failure, SANscape automatically rolls that card's load over to another card.
"We just had a card fail, and SANscape didn't even flinch," says Parker. "It didn't even go down for a second. The rollover happened in a flash. Then SANscape sent out an e-mail that it had lost a card but that everything was fine."
Parker first installed Dot Hill's 4000 Series system on new Dell 8450 servers. The rollout took place over six months. "It only took that long because we were cautious about protecting data before the move," says Parker. "The SAN itself only took two days to set up."
Parker did end up using a clustering solution in tandem with the SAN. Two critical data systems are used for the two different kinds of content that MarketWatch.com provides: news and financial. Today, two main Dell servers take in news content, and two others handle financial content. There's a secondary server for each primary server in a group.
"It's sort of pseudo-clustering," says Parker. "I have two servers on each Dot Hill device. If I lose any one, I'm only losing a single primary."
Parker backs up the company's databases through another RAID device on the same network. Backups are done every night, and the tapes are taken offsite to be stored underground.
Parker has upgraded to Dot Hill's 7000 Series and finds these even faster and easier to expand. The 7000 series offers more than 400MB/second maximum bandwidth, as well as full support for Fibre Channel fabric switching. The upgrade was simplified by the fact that SANnet system uses a shared set of networking components -- including switches, hubs, common hot-swap components, etc. - that support different SANnet generations within the unified SAN.
Dot Hill's service and support made the initial implementation and subsequent upgrade a snap, says Parker. For example, he encountered problems with an I/O channel when deploying the 7000 Series. "The next morning Dot Hill had a new one, and it worked fine," says Parker.
Both the Dot Hill 4000 and 7000 series have delivered on the 99.999% reliability pledge. "These things have been up and running for six months, except when we had a power outage that even wiped out our UPS," says Parker.
The Dot Hill SAN has helped MarketWatch.com live up to its promise of always-available information for its customers. It's also boosted the productivity of MarketWatch.com's bevy of content providers. But Parker is most pleased with another return on the investment: sleep. "I have peace of mind," he says. "I don't go to bed every night wondering what will go wrong."
To find out more about MarketWatch.com, visit its Web site.
To learn more about Dot Hill Systems Corp., visit its Web site.
>>SearchStorage.com Tips: Tech (SAN/NAS)
>>SearchStorage.com Discussion Forums: Storage Networking
>>SearchStorage.com Best Web Links: Storage Area Network (SAN)