MonetDB/X100: a (very) fast column-store

Google Tech Talks
January, 15 2008

ABSTRACT

MonetDB/X100 is a second-generation column-store prototyped at CWI, targeted
at analysis-heavy data management tasks such as data warehousing and
information retrieval.
In this overview talk I will outline the research challenges, novel
techniques and results from this project. The ideal is to close the existing performance gap between hand-coded
applicationas and database systems using architecture-conscious techniques,
and by carefully exploiting the bandwith strengs of modern hardware (and
avoiding its latency weaknesses).
The novel techniques presented encompass expressing relational operators as
vector operations, lightweight compression on the border between CPU cache
and RAM, sequential-access optimized database design and I/O bandwidth
sharing.

Speaker: Peter Boncz

Category:

People & Blogs

Tags:

• google
• techtalks
• techtalk
• engedu
• talk
• talks
• googletechtalks
• education

[davidboreham]

Mmm...real code. It turns out I implemented a similar concept in a project a couple of years ago (in a special purpose database that dealt with airline schedules and fares). Nice to see someone with the brain power to explain the technique clearly. I'm watching, thinking 'ah, so that's what we were doing : avoiding materializing the intermediate results'.

Well worth watching if you're into database architecture.

[OrangutanNationz]

@colinmaharaj what do u mean explicitly?

[chadwickboggs]

After over one year of hands on experience, I confirm that MonetDB is unique, beautiful, and fast. Well done.

[colinmaharaj]

I am trying to figure out something here, how do you explicitly place content in the CPU cache?

[dmpk2k]

Great video. It's a bit hard to understand what the speaker is saying, but the content is completely worth it. I look forward to seeing what else this project achieves.