Parallel Execution in Clarity

The default Clarity task scheduler runs programs in parallel by switching between which protocol is running serially at any time. Protocols are a list of instructions for instruments, which occasionally contain a pause instruction (technically an instance of the ‘’DelayTime’’ class which is derived from the ‘’ProtocolInstruction’’ class). When such an instruction is reached in a protocol (or when the protocol finishes), Clarity assumes all the instruments are in a ready state, and begins executing the next protocol in the queue. Conceptually, this makes the use of automation equipment by different protocols in Clarity similar to a first in first out queue that runs serially. A user can load a protocol, and it will run with exclusive control of the instruments as soon as it is next in the queue and ready to be scheduled.

Although conceptually very simple and robust, by requiring that only one protocol be executing at a time this scheduler may be inefficient. For example, while one instrument is running one task, it might be useful to concurrently use another instrument to perform a task for a different protocol.

However, it is possible to extend Clarity to provide for true parallel execution of protocols, though this is not generally implemented. Clarity’s design assumes that the parallel execution problem for any specific usage scenario will be easier to solve by writing code for an idiosyncratic implementation than specifying that problem for a general framework will be. Clarity provides the tools for a user to relatively easily code a more truly concurrent scheduler and we go over two approaches to implementing truly parallel schedulers here.

Both techniques depend heavily on the ability of virtual instruments to be able to not only modify the instructions of the protocol that called them, but also the entire instrument state. Instructions that do this need only to include the following attributes above the line with the method.

[UserCallableMethod(RequiresCurrentProtocol = true, RequiresInstrumentManager = true)]

With this ability, it is reasonably easy to implement truly parallel schedulers, and we discuss one design patter, the “Parallelism by Dictator” approach that can allow one to do this. However, we emphasize that it will be the responsibility of anyone implementing such custom scheduling operations to ensure that the problems that can arise in parallel computing, such as deadlocks and race conditions, do not occur.

Parallelism by a Dictatorial Virtual Instrument

This approach allows for parallel execution by defining a virtual instrument, the dictator, that acts as a meta-scheduler. It assumes that a lab’s protocols will consist of subsets of instructions that usually perform one of a small number of different task types. For example, a robotic system might be exclusively used to prepare PCR reactions in a liquid handler, or run a bacterial growth curve analysis, or move cells in to new media, but it is not used for other arbitrary activities. Such a small set of tasks might minimally overlap in which instruments they use, and so one can presumably write a quick scheduler that if given a list of protocols could appropriately interweave the PCR reaction tasks with the growth curve analysis tasks.

This could be accomplished relatively easily by having the first instruction in a protocol be to call a virtual instrument and also requests the current instrument manager. This virtual instrument could then examine the system state and decide how to interweave this protocol with the rest. Adjusting the protocols accordingly so that the interwoven events more effectively make use of the instruments.

Clarity by default when running protocol instructions executes atomic instructions and waits for the completion of any one instruction before it resumes execution. To truly implement parallel execution, it is likely best if multiple instruments are simultaneously performing operations. This is relatively easy to do by using the support for Task Parallelism in C#. For example, the code below is a template of a method inside a virtual instrument that can examine the current state and run instructions concurrently.

public bool ExampleParallelRunner(AdditionalMethodArguments eargs)
    //First get the instrument manager
    InstrumentManager IM = eargs.InstrumentCollection;
    //Now we can get all the loaded protocols to look at the system state, perhaps decide what to do
    ProtocolManager pm = Prots = IM.LoadedProtocols;
    //After this we could decide to run two tasks directly with the instruments in parallel
    //First, Grab two instruments, either by asking for it by name or by type
    Twister t = IM.ReturnInstrument("Twister") as Twister;
    TransferStation ts = IM.ReturnInstrumentType<TransferStation>();
    //Now simply perform two methods
    //Remember to add using System.Threading.Tasks to top of the file to use Parallel.Invoke
    //Now we can do two totally random tasks
    Parallel.Invoke(() => ts.CheckIfStationIn());
    return true;

By implementing a virtual instrument with such methods, it can essentially take complete control away from the default scheduler, allowing for the full implementation of a separate parallel machine.