Uppsala research to improve future mobile processors
Future mobile devices will need to store and work with increasingly large amounts of data. Uppsala researchers, together with a world-leading industrial partner, have taken an important step towards making processors more effective at handling these vast data sets.
Information Technology. Photo: Mikael Wallerstedt
“Our approach has become part of the next standard for Arm processors and will lead to significantly improved performance for billions of future mobile and cloud devices,” says David Black-Schaffer, professor in computer science in the Department of Information Technology.
Today’s computer systems face the critical challenge of enabling quick access to enormous quantities of data. Researchers in the Department of Information Technology have now developed a new, more effective method for accessing vast amounts of data in collaboration with Arm, the world’s leading designer of mobile processors.
”Our advance is a new way to effectively store and find the ’translations’ that are needed for every data access. These translations ensure that programs cannot corrupt other programs by accessing their data, but they come at a cost,” says David Black-Schaffer.
Faster data structure searches
Modern operating systems pay a high price for these translations, and applications that access large amounts of data are forced to do many, slow main memory searches to find the right translations. According to David Black-Schaffer, this leads to many programs spending a large portion of their time waiting for the translations, rather than calculating results.
To address this, the Uppsala researchers have developed a new way to store the translations that enables them to search twice as fast. They have accomplished this by reducing the number of levels in the classical tree structure where the translations are stored.
ment of Information Technology. Photo: Private
Their key insight is to change the format of the tree structure to make it possible to use more memory for the tree, but gain performance by reducing the number of steps in the search. But while this sounds simple, the critical contribution is to ensure that the design is compatible with existing operating systems, explains project leader Chang Hyun Park at the Department of Information Technology.
”Working with colleagues at Arm has given us insights into the design challenges we need to address to bring our results into industry. By taking these requirements into account early on, we were able to develop a method that is not just effective but also practical. We are thrilled to see our approach become part of the specification for future processors as a result,” says Chang Hyun Park.
Anneli Björkman