Quintillion started seven years ago with the idea that as the business grew, so would its IT infrastructure. The Irish hedge fund administration company wanted something that could quickly and easily scale, and a virtualized environment with increased server efficiency and easy database deployment seemed like the natural choice. But as the company's data needs grew, its virtualized SQL Server instances hit storage bottlenecks that slowed business performance.
Quintillion was hardly alone among virtualized IT shops when it ran into storage access issues. As applications become virtualized, they can face storage bottlenecks due to slow read/write performance. According to a survey from TechTarget in 2012, 60% of those who had switched from physical to virtual environments were using more storage than they did in physical environments. In the same survey, almost 50% stated they were looking for new tools to manage their virtual machines (VMs).
Charles GillandersCTO at Quintillion
"The nature of the problems we were experiencing was not client-facing," Quintillion CTO Charles Gillanders said. "We continued to meet all of our [service-level agreements], rather we had issues with some of our internal processes."
Gillanders gave an example. When multiple teams were at different stages of work for their clients, they would sometimes experience SQL error 1205, which returns: "Transaction (Process ID) was deadlocked on resources with another process and has been chosen as the deadlock victim. Rerun the transaction." In other words, the server was processing two transactions, but only had the resources for one. So, the SQL Server only proceeded with one of them, notifying the other team to try again later. This meant teams were stuck re-running sometimes lengthy internal processes. This was not good for morale.
Problems also arose when third parties who supplied Quintillion with key information did not send their files for overnight processing. Quintillion's IT and systems developmental staff were able to get the files manually the next morning, but it left them with a 30- to 40-minute wait while the files were being processed. The unexpected additional write load slowed down processing speed and delayed other teams who were trying to extract data from Quintillion's SQL Servers. This led to frustration in the workplace as staff members could not get the performance from their servers that they expected.
Internal processes were one thing, but not being able to meet service-level agreements (SLAs) would be a doubly serious problem for Quintillion, leading to financial penalties from its clients. So it was extremely important for Quintillion to correct the SQL Server problems before they began affecting SLAs.
"The kind of performance issues we were looking to solve had an easy solution -- you throw money at the problem and get additional hardware," Gillanders said.
However, buying a large number of servers is costly and disruptive. As Quintillion continued to grow, it would have to add more and more hardware, each installation temporarily interrupting normal server functioning. Quintillion began engaging with hardware manufacturers, but Gillanders wondered if "software solutions might have better performance and scale more smoothly without having this disruptive cycle."
Serendipity brought Quintillion the answer. Gillanders watched a demo video on YouTube from a company called PernixData. YouTube was not part of the standard process for finding hardware or software upgrades, but PernixData's software appealed to Gillanders. The core of PernixData's philosophy is bringing the data closer to the application, in this case the database. Every step that needs to be taken between the data and the database is one more place a bottleneck could happen. For Gillanders, it made technical sense to bring the cache as close to the server as possible. PernixData's product creates a flash pool from existing, installed servers, and makes it available for all VMs in the data center.
As a small Irish company -- 55 employees, only four of which work in IT and systems development -- Quintillion often has a hard time getting into beta tests of new software in the U.S. However, it had no difficulty getting into PernixData's beta test of PernixData FVP.
"It's a simple solution deployed into old hardware in a short time. That in itself gave me confidence," Gillanders said. The beta test showed immediate improvements and Quintillion became a PernixData customer.
Gillanders said the big win is "not having to dedicate management time to treating some environments differently from everything else." With PernixData's flash pool, no management time has to be taken up with getting SQL Server to recognize or work with PernixData, since PernixData operates outside the servers' awareness. Any acceleration in read/write actions caused by PernixData will register to the SQL Server simply as more speed. Furthermore, With PernixData FVP, Quintillion had no need to dedicate memory spindles.
PernixData FVP's flash pool allows it to perform simultaneous read/write actions across the entire system and can replicate read/write actions to other locations in the database. Gillanders added that PernixData's write caching means Quintillion can access read/write actions rapidly while still using its existing long-term storage area network (SAN). "[PernixData] doesn't replace anything we have today. It just makes what we have work better," Gillanders said.
What Gillanders sees as unique about PernixData is that he does not have to make any changes to the back end or backups. "We have a lot of systems that rely on systems put in place on our SAN," he added. "A simple straightforward solution that enhances rather than replacing is a big win in an environment where switching over the back end would be onerous to achieve."
Gillanders said Quintillion intends to use PernixData in all future database expansions. He said hosting SQL Server in virtual machines affords the company higher availability, improved disaster recovery, ease of deployment, snapshots for tests, replication, business planning, hardware independence, and smoother upgrade and migrations.
"The question isn't, 'Why would you virtualize your SQL Server," he said, "but 'Why wouldn't you virtualize SQL Servers?''