Database Reference
In-Depth Information
Smile: Enabling Easy and Fast Development of
Domain-Specific Scheduling Protocols
Christian Tilgner
1
, Boris Glavic
2
,MichaelBohlen
1
, and Carl-Christian Kanne
3
1
University of Zurich
2
University of Toronto
3
University of Mannheim
{
tilgner,boehlen
}
@ifi.uzh.ch
,
glavic@cs.toronto.edu,
kanne@informatik.uni-mannheim.de
Abstract.
Modern server systems schedule large amounts of concur-
rent requests constrained by, e.g., correctness criteria and service-level
agreements. Since standard database management systems provide only
limited consistency levels, the state of the art is to develop schedulers
imperatively which is time-consuming and error-prone. In this poster, we
present
Smile
(declarative Scheduling MIddLEware), a tool for develop-
ing domain-specific scheduling protocols declaratively. Smile decreases
the effort to implement and adapt such protocols because it abstracts
from low level scheduling details allowing developers to focus on the pro-
tocol implementation. We demonstrate the advantages of our approach
by implementing a domain-specific use case protocol.
1
Introduction
Modern application servers handle large numbers of concurrent requests which
have to be scheduled according to, e.g., correctness criteria like classical serializ-
ability or service-level agreements (SLAs). Standard database management sys-
tems (DBMSs) offer a limited amount of fixed consistency levels, do not provide
sophisticated support for SLAs and, thus, often cannot be used to satisfy domain-
specific scheduling requirements. The state of the art is to develop schedulers
imperatively for applications like Amazon, Ebay or Yahoo [2,5] which yields fine-
tuned schedulers satisfying the application's scheduling constraints. But proce-
dural implementations of schedulers can be complex and dicult to understand,
especially if the request types and correctness criteria are less well studied than,
e.g., classic serializability. Adapting schedulers to evolving requirements results
in expensive and error-prone re-implementations. With our approach we address
these issues by leveraging a declarative language to implement schedulers which
has been shown to be beneficial in previous work [1,3].
1.1 Banking Scenario
We use the following simplified banking scenario to illustrate the shortcom-
ings of standard DBMSs with regard to non-standard scheduling requirements.
A bank institute serves normal and premium customers holding bank accounts.
