We presented our work at VLDB’22
From the 5th to the 9th of September, we joined the VLDB 2022 conference in Sydney, Australia, to present our Scalar DL work. In this blog post, I will briefly introduce the conference, explain what we have presented, and share interesting technology/research directions that I found during the conference.
VLDB
VLDB (Very Large Data Base) is one of the top-tier international conferences in data management and database areas. Unlike common international conferences, VLDB is a place where researchers and practitioners present their papers that are accepted by a scholarly journal called PVLDB. This time was the 48th conference, which took place in Sydney, Australia. The main conference was a 3-day event, and there were also one-day workshops before and after the conference. The conference was sponsored by many tech companies, and we can tell how much the conference has attracted attention from industries.
Scalar DL
As shared in the previous post, our Scalar DL paper was accepted by PVLDB Vol 15, and we had an opportunity to present the work at VLDB’22. We presented it in the last session on the last day of the main conference. The timing was a little unlucky because everyone had already been tired on the last day, but we had many audiences and could appeal our technology to many people (which cannot be seen from the photo, though). Although dealing with Byzantine faults in a transactional database system, especially when it comes to Byzantine fault detection, is a new field of research, we felt it was attracting people’s attention. Please check the following slide deck for the details of the presentation.
Interesting technology directions
I found several interesting technology/research directions during the conference. Let me share some of them briefly.
One interesting direction is an operating system based on distributed DBMSs named DBOS, which is a joint project between universities such as MIT, Stanford, CMU, and UW-Madison. The DBOS project tries to use a distributed transactional DBMS as the basis for a scalable cluster operating system (OS) instead of many instances of a single-node OS like Linux with entirely separate cluster schedulers, distributed file systems, and network managers. Distributed DBMSs naturally have several features that are critical for a cluster OS, such as handling queries and processing transactions at scale. If applications are built on top of a distributed DBMS-based OS, the applications can benefit from them, which could make the applications more simplified and guarantee correctness easily. In VLDB’22, there was one research paper describing the DBOS design and feasibility. There were also related papers that describe the transaction management around DBOS, such as Lotus and Polystore transactions (which was presented at Poly’22 workshop).
Another interesting direction is disaggregated DBMSs. Hardware and software are getting disaggregated for several reasons. Cloud resources are underutilized to accommodate various customers’ requests at various times, so there are many unused dispread resources. Software is also decoupled for simplicity and ease of maintenance. Microservice architecture accelerates the trend and separates a service and a database into multiple single-purpose services and databases. Such disaggregation introduces several challenges: how do we efficiently utilize disaggregated resources and components, and how do we guarantee consistency between multiple disaggregated states, such as the states of multiple heterogeneous databases? In VLDB’22, there were several research papers, such as Redy and Polystore transactions, that address part of the challenges. There was also a paper surveying technologies around microservices and discussing the challenges. Our product Scalar DB also addresses part of the challenges, especially around the data consistency issue.
Summary
We have presented one of our products, Scalar DL, at VLDB’22. It was a great opportunity for us to not only share our work with top-notch researchers and engineers but also grasp technology trends and directions. We would also like to go to VLDB’23 to present our other work.
