Taking SAS® to the server helps research trust achieve new levels of speed and performance
The Nuffield Trust is using SAS® to support essential research projects that influence UK healthcare policy – with a recent move from desktop to server delivering massive gains in processing speed and performance.
An authoritative and independent source of evidence-based research and policy analysis, the Nuffield Trust is helping to drive the debate amongst policymakers, practitioners and the public on the future of health and social care in the UK. The Trust has used SAS® to underpin its analytics work since 2007, with a recent move to the server enhancing its capabilities. "The Trust has a reputation for providing authoritative results people can rely on, and we want to maintain that - which obviously links to our use of SAS," says Adam Steventon, Senior Research Analyst. His team focuses on quantitative analyses involving very large data volumes. "SAS can handle a lot of data, and does so pretty quickly. And the move to the server means it's now astonishingly fast. Even with millions of records involved, individual computations can take just a few minutes."
This work involves large administrative data sets from the NHS and social care sources, including national Hospital Episode Statistics covering 12 million inpatient admissions each year and many more outpatient attendances. "We face technical challenges in how we can store and handle so much data: some projects can have hundreds of millions of records, with multiple sources and formats, from several hundred different providers. We need to clean all that data, and link it together at the person level." Whilst analytical rigour takes precedence over speed, there can also be also benefits gained from the ability to work faster. For example, one part of England might be doing something innovative to prevent hospital admissions, which are undesirable to individuals and costly. It's very useful for the Trust to be able to evaluate outcomes and feedback quickly "to give people on the ground as up-to-date a picture as possible."
From workstation to server
"We wanted to see efficiency gains by changing our set-up," Steventon continues. "We'd started to experience issues with data processing speeds. We'd started running jobs overnight to crunch through the numbers, with computations taking six hours." The team had also started running more ambitious analyses, mainly involving evaluation work, which could take weeks on a workstation simply because so many computations were involved. Another problem resulted from the IT set-up: data flowed from the specialised data server to individual workstations across the intranet, which wasn't designed to handle the volumes of traffic – hundreds of gigabytes of data for analysis - and couldn't move data fast enough. Even though SAS "was doing everything required" the decision was taken to move to another level.
SAS consultants supported the changeover. "I don't think we could have understood by ourselves what the best solution was," says Steventon. "We had meetings with SAS, to think through our requirements and decide on the right solution. SAS provided guidance and support, with consultants spending time here at the Trust to understand what we did and how our technology worked."
More storage space, more power
SAS went live on the server following extensive testing, with the workstations running for a month in parallel during the transition. The quantitative team now has its own dedicated server, with 24 processing units and what Steventon describes as "a large amount of storage space. We fire-up the desktop machines, log-on via remote connection, then run our analyses on the server. We're all using SAS on the server at the same time. It's quick, particularly when we use the server's multi-processing capabilities." Typical work includes working with very large data sets, joining several different data sets, deriving new variables, collapsing data sets - for example, from a hospital admission level to a person level – as well as summarising admissions. "Having built the data we want, we then run a series of analyses, mainly regressions. SAS does all of that very well." His team has the flexibility to put together their own analyses and explore data and results in different ways. "SAS gives us the ability to write our own macros and commands. Most of the analytics are already available in SAS, but some of the more sophisticated ones aren't pre-written functions, so we can write our own," says Steventon.
From two hours to eight minutes
Installing SAS on the server improved processing times dramatically: "We ran tests, and computations that took two hours before took eight minutes on the server, and using just one processor. That was through a combination of changing how we moved data around, made possible by the server, and because the server was simply quicker. We've seen big increases in speed." Other benefits include increased reliability – "The server is far more reliable, and stays on all the time" – and data security. ".." A further benefit is flexibility in terms of working space: "We occupy a listed building, which we love, but it's not ideal as modern office space. Anything that gives us more room is welcome. SAS on the server means anyone can access the system by logging-on remotely, which means more opportunities for hotdesking and flexible working."
He adds, "In general terms, using SAS means we can deal with very large data volumes and it's also quick, crunching through very large data sets in minutes. We're not constrained by time, sitting around waiting for data to be processed."
Copyright © SAS Institute Inc. All Rights Reserved.
The Nuffield Trust
Increase processing speed, efficiency and performance in a highly respected academic research environment; deal with huge data volumes and a wide range of novel data sets, applying a range of proven analytical techniques.
SAS® Analytics, first deployed on the desktop then moved to the server to deliver a faster, more powerful, highly robust and secure solution.
Improvements in speed, efficiency and processing power - some analysis times reduced from two hours to eight minutes; improved data flows, reliability and data security; improved audit, governance and the ability to recreate and share analytical approaches; helping maintain and enhance the Trust's reputation.
“We've seen big increases in speed: some computations that took two hours now take eight minutes on the server. The server is also far more reliable, and we have improved data security. There are fewer data transfers, important given the sensitive nature of our work.”
Senior Research Analyst