Implementation of an affordable computing cluster for pharmacometric analysis
J. G. Coen van Hasselt (1,2), Ron J. Keizer (1,2), M. van Benten (3), Alwin D. R. Huitema (1,2)
(1) Department of Clinical Pharmacology, The Netherlands Cancer Institute, Amsterdam, The Netherlands. (2) Dep. of Pharmacy and Pharmacology, Slotervaart Hospital/Netherlands Cancer Institute, Amsterdam, The Netherlands. (3) Department of ICT, Slotervaart Hospital, Amsterdam, The Netherlands.
Objectives: The aim of this project was to construct a dedicated computing cluster for our population analysis group. The following specifications were defined for this cluster:
- Central installation of software enabling control of integrity and installation;
- access to the cluster from internal and external networks;
- easily extendable with additional nodes;
- use of affordable consumer hardware and preferably open-source software.
Methods: The computing cluster consists of 1 master node and 9 computing nodes. Each node was configured with an Intel quad core CPU allowing execution of 4 processes simultaneously at maximum efficiency. We used Ubuntu Linux Server edition as the operating system. To dynamically distribute computing tasks over the cluster, the software package Sun Grid Engine was used. This package allows efficient distribution of computational tasks across the cluster. Furthermore, a range of applications for pharmacometric data analysis were installed. These included several versions of NONMEM and Fortran compilers, PsN, R and Matlab. The in-house developed and freely available modelling environment Pirana was installed to allow easy access to the pharmacometric software, to offer integrated access to study data, and for processing of results.
User access to the cluster was offered via an SSH connection, both from the internal network and over the internet. Display of graphical interfaces of software that is executed remotely on the cluster, can be easily accomplished using X-forwarding over SSH. As user-side operating system, either Windows, Mac OSX or Linux can be used. The central installation of software on the cluster system enables version control of software, ensuring use of identical versions by all users. Finally, validated scripts, data, models and results can be stored on the cluster with read-only access, improving the regulatory compliance for computerized systems.
Results: The developed computing cluster offers a dedicated and reliable solution for the computational resources needed within our modelling group. Sufficient computing power is available, and this can easily be extended with additional nodes if necessary. Moreover, this system was built using consumer hardware, which makes the system very affordable. Total costs of this system were approximately € 4000,-. The centralized environment in which applications are installed, controlled and executed, increases the integrity of pharmacometric analyses. The installation of Piraña as user interface to the various pharmacometric software, facilitated use of the cluster for both novice and expert users.
Conclusions: This project demonstrates the feasibility of the setup of an affordable and scalable cluster in the pharmacometric setting
 Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN)-a Perl module for NONMEM related programming. Computer Methods and Programs in Biomedicine, Volume 75 (2004), Issue 2, Pages 85-94
 R Development Core Team (2009). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.r-project.org/.
 Keizer RJ, van Hasselt JGC, Huitema ADR. A simple infrastructure and graphical user interface (GUI) for distributed NONMEM analysis on standard network environments. PAGE 19 (2010) Abstr 1689 [www.page-meeting.org/?abstract=1689]