me

Marcin Żukowski

contact vectorwise research publications talks varia

Contact

To contact me, please write to marcin at vectorwise dot com.

VectorWise

In 2008 the MonetDB/X100 project described further on this page has been converted into VectorWise, a company which I currently manage.

VectorWise cooperates with Ingres, the leading open-source database company, on the Ingres VectorWise product.

Research

Between 2001 and 2008 I was a member of the MonetDB development team. My initial work with MonetDB concentrated on parallel query execution, which resulted in my master's thesis (pdf).

From 2003 I was working on the MonetDB/X100 project under the supervision of Peter Boncz and Martin Kersten.

This project resulted in my PhD thesis and was a base for the VectorWise company

MonetDB/X100 [DEBull05, VLDB05] introduced a novel query processing architecture, based on the concept of "vectorized in-cache processing" [CIDR05]. Comparing to the traditional tuple-at-a-time processing model, it significantly reduces the interpretation overheads and makes most of the CPU time go into the actual data processing. Additionally, this architecture allows efficient use of modern processing hardware [CIDR05]. The project resulted in various innovations in the area of query execution, including high-efficiency hash-tables [DAMON06], novel approach to in-flight data storage [DAMON08], and a new method of handling updates [SIGMOD10] In-memory, X100 can provide even ca. 1GB/s data processing throughput on a single CPU. This high performance calls for high-bandwidth storage facility. Here, X100 uses a column-based storage, that introduces two bandwidth-improving techniques [BNCOD05]. "Lightweight in-cache compression" [ICDE06] allows trading some processor time for an increased perceived disk bandwidth. "Cooperative scans" [VLDB07] dynamically schedule query activity and I/O operations to amortize the cost of disk access among multiple queries. As a result, X100 has been shown to provide very high performance both for in-memory [CIDR05] and disk-based [ICDE06] database tasks, as well as in large-scale information retrieval scenarios [TREC06, CIDR07, VLDBJ08]. Additionally, the ideas behind MonetDB/X100 were influential for a prototype query execution kernel for STI Cell processors [DAMON07].

Publications

Talks

Varia

Valid HTML 4.01! Valid CSS!