Guidelines for selecting appropriate classes of algorithms. The KernelMerger tool is a newly developed tool, which especially in the QualiMaster adaptive pipeline project proves that it can be highly effective. The question to be answered is: What is the kind of algorithms that would best take advantage of the tool? This question has many answers, not only in terms of the characteristics themselves, but also on the type of applications which to date cannot be executed on any platform of any vendor, and which this tool will enable. To illustrate, let’s assume that we have 2-3 algorithms, which –depending on the application characteristics- are adaptively switched in and out of the custom hardware. To date there are only two ways in which this can be achieved: by reconfiguring the device in between algorithms, and by having some version of all of the algorithms inside the reconfigurable fabric, so that selection of one of them is merely a matter of routing the data to the appropriate module. Both of the above approaches do have applications to date. We will give examples outside the realm of QualiMaster in order to emphasize the general usefulness of the KernelMerger tool. If, for instance, we have a network processor which checks for denial-of-service attacks, a periodic update of the patterns might be done daily or weekly, and in this case reconfiguring the custom hardware might have an acceptable overhead, as some microseconds (or even milliseconds) for reconfiguration is an acceptable tradeoff. For other applications, however, e.g. weak vs. strong encryption for the real-time processing of data might necessitate for all encryption algorithms to be on the reconfigurable fabric at all times, as the switch among them might be needed right at the time of en emergency situation, and thus there is no time for the reconfiguration. This latter case means that vast hardware resources remain idle at all times, and the cost of hardware is for much larger devices vs. what is actually needed at any given time. The KernelMerger tool solves this problem by reusing resources among different applications, as long as not all of them are required at the same time. Case-in-point for the QualiMaster project, was that as described in this deliverable, one could have either multiple algorithms at the same time with fewer resources than the sum total of the resources for each algorithm, or, have differently dimensioned versions of the same algorithm all residing on the hardware, so that the correct version will be available at all t...
