Toward high performance parallel experimentation machines. Use of a scheduler as a quantitative computer-aided design tool for evaluating workstation performance

Abstract Parallel experimentation machines offer the opportunity to acquire large volumes of scientific data in short periods of time. However, envisioning the performance of the integrated assembly of hardware components (robot, sample handling modules, analytical instruments) is a stubborn and pervasive design problem in the field of laboratory automation. An automated chemistry workstation has been built that consists of discrete hardware modules and is equipped with an experiment manager software package; the latter includes a scheduler for performing parallel experiments. In this paper, the scheduler is used as a computer-aided design (CAD) tool to evaluate the effects of speed enhancements in various hardware modules, including robot translation rate, sample manipulator syringe speed, solvent dispensing with a high-speed pump, and replacing traditional batch methods of sample analysis with flow injection analysis. The proposed hardware alterations are feasible with existing technology. The CAD approach has three elements. First, each alteration is evaluated quantitatively using timing formulae containing variable speed parameters. Second, the overall system performance improvement due to each individual alteration is simulated by scheduling a typical workload of experiments. Third, performance is equated with the number of experiments that can be performed in parallel. The relative importance of the hardware alterations depends on the typical workload of experiments. A workstation that incorporates all alterations exhibits a performance improvement of five to fifteenfold compared with our existing workstation (and over 150-fold compared with serial operation). Such a workstation would be capable of initiating and monitoring > 300 reactions (16 h each) in parallel, a throughput corresponding to ≈ 1700 experiments per week. Application of the scheduler as a CAD tool thus provides a means to predict performance quantitatively. This approach should prove generally useful in laboratory automation and is crucial for guiding the systematic design of high performance parallel experimentation machines.