FAS a Freshness-Sensitive Coordination Middleware
for a Cluster of OLAP Components
Uwe R6hm
Klemens B6hm*
Hans-J6rg Schek
Swiss Federal Institute of Technology
ETH Zentrum, 8092 Zurich, Switzerland
{ roehm,boehm,sche k,schuldt } @ in f.ethz.ch
Heiko Schuldt
1 Introduction
Data warehouses offer a compromise be-
tween freshness of data and query evalua-
tion times. However, a fixed preference ra-
tio between these two variables is too undif-
ferentiated. With our approach, clients sub-
mit a query together with an explicit
ness limit
as a new Quality-of-Service pa-
rameter. Our architecture is a cluster of
databases. The contribution of this article
is the design, implementation, and evalua-
tion of a coordination middleware. It sched-
ules and routes updates and queries to clus-
ter nodes, aiming at a high throughput of
OLAP queries. The core of the middleware
is a new protocol called FAS
Aware Scheduling)
with the following qual-
itative characteristics: (1) The requested
freshness limit of queries is always met,
and (2) data accessed within a transaction
is consistent, independent of its freshness.
Our evaluation shows that FAS has the
following nice properties: OLAP query-
evaluation times are close (within 10%) to
the ones of an idealistic setup with no up-
dates. FAS allows to effectively trade 'up-
to-dateness' for query performance. Even
when all queries request fresh data, FAS
clearly outperforms synchronous replica-
tion. Finally, mean response times are inde-
pendent of the cluster size (up to 128 nodes).
* Current affiliation: Otto-von-Guericke-Universit~it Magde-
burg, Germany
Permission to copy without fee all or part 03" this material is
granted provided that the copies are not made or distributed ]'or
direct commercial advantage, the VLDB copyright notice and the
title of the publication and its date appear, and notice is given that
Data warehouses are closely tied to OLAP, i.e., on-
line analytical processing of the vast amount of data
of an organization. They typically offer a compro-
mise between freshness of data and warehouse main-
tenance costs. Different application scenarios and
users however have different preferences in this re-
spect, and a fixed preference ratio is too undifferen-
tiated. With our approach, clients submit a query
together with an explicit
freshness limit
as a new
Quality-of-Service parameter. In other words, read-
ers may decide infinitely variable how much up-to-
date the data accessed should be. The goal is to
use this additional information to improve through-
put. The concern of this article is the through-
put of a stream of OLAP queries, i.e., we assume
a read-mostly environment with many concurrent
readers. This complements recent work on replica-
tion in OLTP scenarios, e.g., [6, 17].
The object of this study is a cluster of databases
[18, 14]: this is a cluster of commodity computers,
each node running an off-the-shelf database manage-
ment system as transactional storage layer. This pa-
per assumes that all cluster nodes are homogeneous,
i.e., they run the same DBMS with the same database
schema. Each node holds a full copy of the database,
but the freshness of these copies may vary between
cluster nodes. Finally, we assume that there is a
coordination middleware layer on top of the clus-
ter (cf. Figure 1). Clients submit query or update
transactions to this middleware, instead of directly
communicating with specific cluster nodes. The mid-
dleware schedules and routes updates and queries to
cluster nodes. The
generates a correct in-
terleaved execution order. In general, scheduling al-
lows for several cluster nodes where a query may ex-
ecute. The
chooses one of these nodes for each
copying is by permission of the Very Large Data Base Endowment.
To copy otherwise, or to republish, requires a fee and~or special
query. The challenge with such a cluster architecture
permission from the Endowment.
is (1) to achieve high performance with regard to the
Proceedings of the
28th VLDB Conference, OLAP query stream, (2) to guarantee global correct-
Hong Kong, China, 2002 ness at the same time, and (3) to satisfy the freshness

Get Proceedings 2002 VLDB Conference now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.