Queuing Question
Added by Anonymous almost 15 years ago
In looking at the differences between the breadth and depth queuing options, I ended up wondering which one was faster (and what the pros and cons were for each one). More specifically, I was wondering what the major advantage of using breadth-first tree-reversal as opposed to using depth-first tree reversal was; with breadth-first, you seem to get more problems due to hysteresis than you seem to eliminate by not having to switch the magnitude as often.
Somewhat related: does using the Queuing method take longer than doing the image acquisition 'manually'? The manual mentioned that constant interaction with the drift manager was something to watch out for since it would increase processing time considerably...
Replies (2)
Re: Queuing Question - Added by Anchi Cheng almost 15 years ago
Breadth-first mode is slower than Depth-first if both are fully automated for exact reason you stated. The additional drift management need makes it so. The advantage of Breatdth-first mode comes in when it is not fully automated. Consider the situation that you have very few particles per hole or they distribute rather unevenly in a way that the target finder can not choose the good hole with good particles and ice over the holes that are useless, but a human can. Each final acquisition, including focus and image acquisition may take about 30-60 sec (depending on the camea speed).
Let's say that you want 1200 good images and can only get 50% of them by full automation with Depth-first, then you will have to waste 1200 x 30 sec = 10 hr. The total Leginon session time after setup is (30+20) * 1200 + 10 hr = 26.6 hr. The 20 sec is the time for Leginon to move and acquire hl image drift management etc.
If you choose the good holes yourself manually in Depth-first, you will save almost all this 10 hr time. However, you will have to stay with the computer, picking the target every 30+20+5 sec. The 20 sec is the time for Leginon to move and acquire hl image for you to pick on, and the 5 sec is the time you decide where to pick the targets. In total, you will sit there for 18.3 hr straight while the Leginon session takes the same amount of time acquiring images. The only gain is that you won't be doing the focusing manually.
If you choose the good holes manually in Breadth-first mode, you can pick the 1000 targets while Leginon acquires the hl images. You probably sit there for 5 sec * 1200 = 1.6 hr picking the targets. Then you start the queue process and come back every two hours to fill the liquid nitrogen. Let's say that the extra drift management takes 15 sec per target (usuaully less), Your Leginon session is now (30+20+15) = 21.6 hrs. However, your personal time is 1.6 hr.
Therefore, personal time is what Breadth-first mode saves. The larger the dataset you need, the more important this is for you, right?
Anchi
Re: Queuing Question - Added by Anonymous almost 15 years ago
Ah, thank you for clarifying this! It was really helpful, and I think it will help us decide how to proceed today on the microscope.