Questions and Answers :
Wish list :
Distinct applications instead of 'boinc2docker'
Message board moderation
Author | Message |
---|---|
Send message Joined: 29 Dec 18 Posts: 7 Credit: 13,559 RAC: 0 |
On your website it reads as follows:
When taking a look at the client_state.xml file, only ONE application name is defined. I am hereby asking you to re-organize the way you deploy apps. And for this request, I have a number of good reasons: 1. When first trying to contribute to this project, I realized that running more than one task on a Windows system with 8 GB of RAM will result in a reproducible freezing/bluescreen of the machine. Obviously, your DC project by-passes the BOINC-implemented memory management, which is not unusual for Virtualbox-based projects (such as our own, too). Meaning in practice: People coming to your project using standard settings will not stay for long unless they have machines with non-standard consumer RAM equipment. I solved this issue by allowing only 1 nanoHub task at a time - which, however, requires people to be firm with app_config. xml settings. Newbees won't manage that. Result: You loose valubale contributors / compute power. 2. The apps are declared as (mt), i.e. capable of multithreading. Well, I do not see that any of the apps is indeed mt-capable. Instead, each task tested uses at maximum a single core while the rest are idle. Still these tasks run around 1-2 minutes at maximum. Meaning in practice: A massive waste of compute possibilities. Heavy data transfer frequency. If you declare each of the (as you say) >200 apps as separate entities, it will become possible for users to fine tune settings for each of the apps - with respect to RAM, run times, combination with each other and with other BOINC projects, etc. You could declare the maximum system requirements for each of these apps on your website such that people see which of the apps are e.g. RAM hungy and which are not. I for example prefer apps that run for long, so please also consider bundling of multiple tasks of the same type into one task (examples can be retrieved from other DC projects, e.g. Milkyway@home). Michael. President of Rechenkraft.net e.V. - Germany. |
©2025 COPYRIGHT 2017-2018 NCN