feedbacklog.typ

#import "/layout/feedbacklog_template.typ": *
#import "/metadata.typ": *
#import "/utils/feedback.typ": *

#set document(title: titleEnglish, author: author)

#show: feedbacklog.with(
  titleEnglish: titleEnglish,
  examiner: examiner,
  supervisors: supervisors,
  author: author,
  presentationDate: presentationDate,
  feedbacklogSubmissionDate: feedbacklogSubmissionDate,
)

#feedback(
  feedback: "Since the attention span of the audience is rather short, try getting to the problem as fast as possible. Several slides in the presentation display redundant or repetitive information and could be deleted",
  response: "To  get  to  the  problem  earlier  in  the  presentation  as  well  as  to  present  only relevant  information,  some  of  the  slides  are  removed  (Agenda)  and  some  of  them  are  merged (Context, Motivation and Objectives), therefore the “Problem” slide is now at slide 4.",
)

#feedback(
  feedback: "The structure of the presentation is rather unique; it would be great if you could help the audience to follow it easier.",
  response: "To help the audience to follow the presentation easier, a progress-bar was added at the bottom of the slides to show the milestones of the presentation and highlight the one that indicates the current slide.",
)

#feedback(
  feedback: "Is the resource pool, as the main contribution, independent of the programming exercise language?",
  response: "Yes, the resource pool is completely independent of the exercises.",
)

#feedback(
  feedback: "The pool does not load the JDK etc., since prebuilt IDE images are used instead. However, the pool still thinks of images differently, so isn't there a dependency after all?",
  response: "There is a dependency in that sense, as raised by Prof. Krusche.",
)

#feedback(
  feedback: "Was the benchmark done with only one specific blueprint?",
  response: "Yes, the benchmark was only done with one version/language configuration.",
)

#feedback(
  feedback: "Can you explain how the resource pooling works? What happens if there are many different configurations?",
  response: "Prewarming happens separately, so there is a pool for each version. Different versions are unrelated to each other in that sense.",
)

#feedback(
  feedback: "How many prewarmed resources were there for the benchmark?",
  response: "For the benchmark with 50 concurrent users, there were 50 prewarmed resources. If demand exceeds that, the system falls back to lazy provisioning.",
)

#feedback(
  feedback: "Do you think it is realistic to have 50 prewarmed sessions?",
  response: "Yes, the system can definitely support over 100 without any issues, so this is realistic.",
)

#feedback(
  feedback: "Are prewarmed resources created at the same time? Does prewarming happen automatically in the background?",
  response: "Predictive scaling would ideally determine the size of the pool automatically, but currently this is done manually.",
)

#feedback(
  feedback: "How many resources does one session contain?",
  response: "0 CPU and a small amount of RAM.",
)

#feedback(
  feedback: "So the prewarmed resources basically sleep until needed - is waking them up faster?",
  response: "Yes, since it is just an HTTP service that isn't being used, waking it up is fast.",
)

#feedback(
  feedback: "The presentation was good overall, with mostly dynamic behavior and a good visual representation. Matthias really liked the thesis and the work done.",
  response: "",
)

#feedback(
  feedback: "(This is about future ideas) How exactly do you determine how many resources you want to provision?",
  response: "Deterministic data is used first, such as exam schedules, since there is no problem having too many resources. These simple signals are taken into account first.",
)

#feedback(
  feedback: "(This is about future ideas) Is this deterministic data provided by the instructors?",
  response: "No, the data is taken directly from Artemis, without any manual intervention from the instructors.",
)