Working alongside the Professor Willy Zwaenepoel, Dean of the Faculty of Engineering, and computer science researcher Dr Florin Dinu, the research paper focused on developing a method to overcome latency spikes.
The large-scale storing and processing of online data is a challenge faced by most organisations, and its importance is only set to rise as our professional and personal lives are increasingly experi-enced online.
Logging into an online account, downloading a file from a cloud server, watching a video on YouTube or liking someone’s social posts make up just a fraction of the millions of requests large online databases, such as Google, iCloud and Facebook, answer each second.
Traffic increases mean that online servers must play catch up, recalibrating and optimising their response times. The researchers’ work aims to decrease such response times, that are commonly slowed due to poor sharing of computing resources.
“Just like driving in peak hour, a spike in traffic can cause a bottleneck of requests. We’ve all experienced a slow website — often the cause is an inundation of requests — we call this an increase in latency,” explained Oana.
Professor Zwaenepoel emphasised that there is a growing need for technology that effectively deals with surges in traffic — a need that will only become more pronounced as the online world becomes the centre point for most commercial activity.
“As more people come online, large online servers are dealing with an increasing level of traffic and requests each day. Our method focuses on generating shorter response times for customer requests to ensure performance predictability and improve user experience,” said Professor Willy Zwaenepoel.
“This reliability requires complex infrastructure, however the issue is large online services need to create a predictable experience while dealing with large swathes less-predictable requests."
According to Oana, online systems such as Facebook, Google and iCloud are experiencing a period of rapid expansion. By providing lower latency and better predictability, she believes her research will improve the reliability of data transmission.
“Improvements might be in seconds or milliseconds, however, this all adds up. Improved latency contributes to greater productivity and a better user experience,” said Oana.
“We have designed a scheduler that ensures the computing resources are split productively between the customer requests and routine maintenance work. Our solution ensures that customer requests are never blocked by maintenance work,” she said.
Usenix ATC is one of the industry’s largest conferences, attended by researchers from across the globe.
“This was our very first appearance at Usenix so it’s remarkable that we have won the best paper award!” Professor Zwaenepoel concluded.
The research has been conducted in collaboration with Silicon Valley software company, Nutanix. The code used for the paper is open-source.