Skip to main content
To KTH's start page To KTH's start page

Making data accessible - despite server and network issues

Published Nov 22, 2023

A software developed by KTH has enabled cloud services to be easier to build and more resilient to failures.

“I wanted to make fault-tolerant systems easy to build and use, contrary to what is often perceived of such systems,” says Harald Ng, PhD student at KTH.

Social media, photo albums, and digital wallets are examples of applications that from the user’s perspective always appear available. These applications runs on cloud services that need to be correct even when failures occur on the network and servers. The challenge of creating something that allows correct data access despite network and server failures has long intrigued Harald Ng, a PhD student at KTH. Together with professors Paris Carbone and Seif Haridi, they have developed OmniPaxos, a software library capable of handling complex network problems and ensuring correct access to data .

In simple terms, the software ensures that data is accessible by putting it on multiple servers. Even if some servers crash or disconnect from each other, the software ensures that the data you access is consistent, i.e. the same.

"There were similar systems before, but what is unique in what we have created is that OmniPaxos is particularly good at handling partial network failures,” says Harald Ng.

Harald Ng, PhD student KTH

Failures occur more often

Partial network failures imply that different system servers have different views of the network, which makes them unable to reach agreement and prevents data from being replicated properly.

"We discovered that today's network infrastructure is often virtual and software-based, meaning that partial network failures occur more often,” he says.

A few months ago, the study was presented at EuroSys, a premier venue for systems research. Since then, much time has been spent on making OmniPaxos available as a simple software. The goal is to make building cloud services as simple as building a local application. This has also been a collaborative effort with the students from the advanced Distributed systems course given at KTH.

”The driving force has been that many people have considered systems like this to be complicated, while we had the idea that it doesn't have to be so difficult,” he says.

So far, 1 400 people have downloaded the software, and the hope is that it will be used in larger commercial and open-source systems. Further work is underway to make OmniPaxos even more efficient and adapt to different situations.

”Of course, we hope that bigger companies and systems will pick it up. We have seen from talks and posts that they are often looking for solutions like this," he concludes.

Emelie Smedslund

Related news

  • No up-to-date news right now.