PhD Course - Database Systems on Modern Hardware
Database Systems on Modern Hardware
Philippe Bonnet, Pinar Tözün, Björn Thor Jónsson
Date(s) of the course
09-11/2018 (10 sessions)
This course will focus on techniques developed in recent years so that database systems can take advantage of modern hardware (multicore, large RAMs, fast NAND, RDMA).
- LSM Tuning: Dostoievsky Sigmod18, Triad ATC17
- Buffer management: LeanStore ICDE18, Siberia VLDB14
- DB on Manycores: Morsel-driven parallelism SIGMOD14, BionicDB
- Query compilation: Compiling Query Plans VLDB11, Relaxed Operator Fusion VLDB18
- SSD failures: SSD Failures in Datacenters Systor16, Reliability of NAND-based SSDs IEEE17
- NVM: NVM in DBMS Sigmod18, NVMRocks ISTC17
- Indexes: BzTree VLDB 2018, Building a Bw-Tree Sigmod18
- Concurrency Control: Feral CC Sigmod15, Empirical study of MVCC VLDB17
- Networked DB: Clusters of HW Islands Damon16, Rack-scale Joins Sigmod15
The course is organized in 9 sessions. In each session, two papers are read and discussed. The objective is that in each session, discussions lead to derive from the papers the main insights, key techniques and open issues.
Yes. Students should be familiar with database internals. They should have taken a MSc-level course on database systems.
2 presentations during the course of the class and a report based on the summaries of the 9 sessions.
As a rule of thumb, courses of 7.5 ECTS or more must be approved as MSc courses.
Amount of hours the student is expected to use on the course
Participation: 9x3 hours
Preparation: 9x5 hours
6, 3 from ITU 3 from DIKU