On Thu, Sep 10, 2009 at 12:14:29AM -0400, Luis Useche wrote: > Hi Guys, > > I am having some time free soon and I feel like doing some hacking in > the OpenBSD kernel. I would like to work in the I/O stack. I would > prefer something easy to do to get introduced to the kernel. I was > thinking on implementing a simplified version of FS2 > (http://citeseerx.ist.psu.edu/showciting?doi=10.1.1.125.4078). This > project would increase performance and reduce power consumption. I > have a couple of question: > > 1. I would like to know if that project is interesting to the > community. After all, I would prefer to spend my time in something > useful. If not, any other suggestion for project?
I think that the assumptions made by the FS2 paper are outdated. Modern systems have often either SSD or good raid controllers with large buffers in them. In both cases the "seek" time is unpredictable. Additionally modern disk can re-map blocks and so it is not known if two blocks are close by or not. So in the end I think the amount of code needed to track these additional copies of blocks and figuring out if they're still valid or not will bloat the code without big benefit. If you like filesystem hacking there are a few projects in other BSD projects worth looking at. e.g. NetBSD has a journaling mechanism for FFS. Another thing is a good mfs replacement or a FS suitable for flashes. In the end it is your hacking time so you need to choose something that realy bugs you and is not a huge monster project. > 2. Any suggestion to get introduced to the kernel easily? Any advice? > I already have the "4.4BSD design and implementation" book. > Reading source code and find a problem you like to get solved because it bugs you very much. Additionally there are some kernel API man pages and google can help as well in some cases. -- :wq Claudio
