London-based TrafficBroker is an online agency whose main business apps are IOPS-hungry, homegrown advertising search keyword management tools, reporting databases and Web servers.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
The server virtualisation project saw seven servers with direct-attached storage (DAS) reduced to three 16-core Dell machines with 60 virtual servers. The project was driven by the need to support high availability and the rapid development of new applications for clients in a business whose size had doubled in a year.
"We couldn't keep running our data centre the way we used to," said Lukas Oberhuber, TrafficBroker's chief technology officer. "It wouldn't have worked. We had a hodgepodge of incompatible servers that were purchased at various times and with varying power, and any time we wanted to do something new it was extremely painful and soaked up weeks of time we couldn't afford."
The product implemented was a 3PAR InServ storage array, initially with just one tier comprising 12 TB of 7,200 rpm SATA drives. This was later uprated to 24 TB of capacity in two tiers with 15,000 rpm Fibre Channel (FC) drives added to provide much higher availability levels than first calculated. Data is manually directed between the two tiers -- production apps and databases on tier 1 and test, and development and lower priority data on tier 2.
Oberhuber explained the thinking behind moving from one tier to two.
"We started with just 12 TB of SATA. At first the data wasn't tiered at we had calculated -- we wanted speeds of 1,000 IOPS, but soon realised we needed 2,500 IOPS," he explained. "The database servers get hammered really hard loading in reports, and the VMs [virtual machines] were competing for IOPS."
Cost of adding drives the winning factor3PAR won out based on the long-term cost of ownership, specifically the cost of adding extra drives, Oberhuber said.
"In the end, it was a bake-off with Compellent," he said. "3PAR offered a better pricing structure over the long haul and the references were very strong. The key cost consideration in the long term was the price of adding drives, and it's in this that lots of vendors bite you in the tail. Compellent had more bells and whistles, such as dynamic tiering, but we felt we didn't need those extra features at the time."
As a result of the implementation of shared storage, TrafficBroker has made savings in staff time and hardware purchases. Data storage administration is far easier with all disk pooled in the 3PAR storage-area network (SAN), while thin provisioning has allowed TrafficBroker to safely overallocate virtual capacity up-front without having to purchase physical capacity until it's actually needed for written data.
Oberhuber said his staff now spends 33% less time dealing with storage capacity issues than before the 3PAR implementation. Capacity usage has been boosted by not having disk on separate servers, while thin provisioning has helped reduce physical capacity requirements by 25%.
Separating storage from servers has allowed the business to scale without disruption, and to eliminate planned downtime and the need to take applications offline.
The key benefits, Oberhuber said, have been to allow the business to grow rapidly and satisfy high demand from developers on the server and storage environment.
"It wasn't an exercise in simply making savings. We wouldn't have been able to cope with a business that's doubling in size every year without virtualising servers and implementing shared storage. Our developers need to get new apps up and running in a day, and you need the shared storage to support the VMs for that," he said.
When asked what 3PAR could do to improve future versions of its products, Oberhuber said he'd like to see some of the advanced features other vendors offer.
"I think some bells and whistles on the 3PAR product would be good," he added. "Dynamic tiering would be interesting. And we'd like to be able to monitor usage and know when we're hitting the limits of IOPS primarily, as well as disk capacity."